blob: 7d424226756537ab6313ff5004ed83a19b6fc059 [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * Health-checks functions.
3 *
Willy Tarreau26c25062009-03-08 09:38:41 +01004 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02005 * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl>
Willy Tarreaubaaee002006-06-26 02:48:02 +02006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 */
13
Willy Tarreaub8816082008-01-18 12:18:15 +010014#include <assert.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020015#include <ctype.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020016#include <errno.h>
17#include <fcntl.h>
Willy Tarreau9b39dc52014-07-08 00:54:10 +020018#include <signal.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090019#include <stdarg.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020020#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020022#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020023#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <unistd.h>
25#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040026#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090027#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020028#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020029#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020030#include <arpa/inet.h>
31
Willy Tarreauc7e42382012-08-24 19:22:53 +020032#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020033#include <common/compat.h>
34#include <common/config.h>
35#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020036#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020037#include <common/time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020038
39#include <types/global.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090040#include <types/mailers.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020041#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010042#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020043
44#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020045#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010046#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020047#include <proto/fd.h>
48#include <proto/log.h>
49#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020050#include <proto/port_range.h>
Willy Tarreau3d300592007-03-18 18:34:41 +010051#include <proto/proto_http.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010052#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010053#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010054#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020055#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020056#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010057#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020058#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020059#include <proto/log.h>
60#include <proto/dns.h>
61#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020062
Willy Tarreaubd741542010-03-16 18:46:54 +010063static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090064static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020065static char * tcpcheck_get_step_comment(struct check *, int);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +020066static void tcpcheck_main(struct connection *);
Willy Tarreaubd741542010-03-16 18:46:54 +010067
Simon Horman63a4a822012-03-19 07:24:41 +090068static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010069 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
70 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020071 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020072
Willy Tarreau23964182014-05-20 20:56:30 +020073 /* Below we have finished checks */
74 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010075 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010076
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010077 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020078
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010079 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
80 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
81 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020082
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010083 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
84 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
85 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020086
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010087 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
88 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020089
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020090 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
93 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
94 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +090095
96 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
97 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +020098 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020099};
100
Cyril Bontéac92a062014-12-27 22:28:38 +0100101const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
102 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
103 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
104 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
105 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
106 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
107 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
108 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
109 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
110 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
111 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
113};
114
Simon Horman63a4a822012-03-19 07:24:41 +0900115static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100116 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
117
118 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
119 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
120
121 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
122 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
123 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
124 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
125
126 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
127 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
128 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
129};
130
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200131/*
132 * Convert check_status code to description
133 */
134const char *get_check_status_description(short check_status) {
135
136 const char *desc;
137
138 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200139 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200140 else
141 desc = NULL;
142
143 if (desc && *desc)
144 return desc;
145 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200146 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200147}
148
149/*
150 * Convert check_status code to short info
151 */
152const char *get_check_status_info(short check_status) {
153
154 const char *info;
155
156 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200157 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200158 else
159 info = NULL;
160
161 if (info && *info)
162 return info;
163 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200164 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200165}
166
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100167const char *get_analyze_status(short analyze_status) {
168
169 const char *desc;
170
171 if (analyze_status < HANA_STATUS_SIZE)
172 desc = analyze_statuses[analyze_status].desc;
173 else
174 desc = NULL;
175
176 if (desc && *desc)
177 return desc;
178 else
179 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
180}
181
Willy Tarreaua150cf12014-05-20 21:57:23 +0200182/* Builds a string containing some information about the health check's result.
183 * The output string is allocated from the trash chunks. If the check is NULL,
184 * NULL is returned. This is designed to be used when emitting logs about health
185 * checks.
Willy Tarreauddd329c2014-05-16 16:46:12 +0200186 */
Willy Tarreaua150cf12014-05-20 21:57:23 +0200187static const char *check_reason_string(struct check *check)
Willy Tarreauddd329c2014-05-16 16:46:12 +0200188{
Willy Tarreaua150cf12014-05-20 21:57:23 +0200189 struct chunk *msg;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200190
Willy Tarreaua150cf12014-05-20 21:57:23 +0200191 if (!check)
192 return NULL;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200193
Willy Tarreaua150cf12014-05-20 21:57:23 +0200194 msg = get_trash_chunk();
195 chunk_printf(msg, "reason: %s", get_check_status_description(check->status));
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200196
Willy Tarreaua150cf12014-05-20 21:57:23 +0200197 if (check->status >= HCHK_STATUS_L57DATA)
198 chunk_appendf(msg, ", code: %d", check->code);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200199
Willy Tarreaua150cf12014-05-20 21:57:23 +0200200 if (*check->desc) {
201 struct chunk src;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200202
Willy Tarreaua150cf12014-05-20 21:57:23 +0200203 chunk_appendf(msg, ", info: \"");
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200204
Willy Tarreaua150cf12014-05-20 21:57:23 +0200205 chunk_initlen(&src, check->desc, 0, strlen(check->desc));
206 chunk_asciiencode(msg, &src, '"');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200207
Willy Tarreaua150cf12014-05-20 21:57:23 +0200208 chunk_appendf(msg, "\"");
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200209 }
210
Willy Tarreaua150cf12014-05-20 21:57:23 +0200211 if (check->duration >= 0)
212 chunk_appendf(msg, ", check duration: %ldms", check->duration);
213
214 return msg->str;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200215}
216
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200217/*
Simon Horman4a741432013-02-23 15:35:38 +0900218 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200219 * an adequate CHK_RES_* value. The new check->health is computed based
220 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200221 *
222 * Show information in logs about failed health check if server is UP
223 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200224 */
Simon Horman4a741432013-02-23 15:35:38 +0900225static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100226{
Simon Horman4a741432013-02-23 15:35:38 +0900227 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200228 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200229 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900230
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200231 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100232 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900233 check->desc[0] = '\0';
234 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200235 return;
236 }
237
Simon Horman4a741432013-02-23 15:35:38 +0900238 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200239 return;
240
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200241 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900242 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
243 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200244 } else
Simon Horman4a741432013-02-23 15:35:38 +0900245 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200246
Simon Horman4a741432013-02-23 15:35:38 +0900247 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200248 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900249 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200250
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100251 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900252 check->duration = -1;
253 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200254 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900255 check->duration = tv_ms_elapsed(&check->start, &now);
256 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200257 }
258
Willy Tarreau23964182014-05-20 20:56:30 +0200259 /* no change is expected if no state change occurred */
260 if (check->result == CHK_RES_NEUTRAL)
261 return;
262
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200263 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200264
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200265 switch (check->result) {
266 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200267 /* Failure to connect to the agent as a secondary check should not
268 * cause the server to be marked down.
269 */
270 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900271 (check->status >= HCHK_STATUS_L57DATA)) &&
Willy Tarreau12634e12014-05-23 11:32:36 +0200272 (check->health >= check->rise)) {
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200273 s->counters.failed_checks++;
274 report = 1;
275 check->health--;
276 if (check->health < check->rise)
277 check->health = 0;
278 }
279 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200280
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200281 case CHK_RES_PASSED:
282 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
283 if ((check->health < check->rise + check->fall - 1) &&
284 (check->result == CHK_RES_PASSED || check->health > 0)) {
285 report = 1;
286 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200287
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200288 if (check->health >= check->rise)
289 check->health = check->rise + check->fall - 1; /* OK now */
290 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200291
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200292 /* clear consecutive_errors if observing is enabled */
293 if (s->onerror)
294 s->consecutive_errors = 0;
295 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100296
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200297 default:
298 break;
299 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200300
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200301 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
302 (status != prev_status || report)) {
303 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200304 "%s check for %sserver %s/%s %s%s",
305 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200306 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100307 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100308 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200309 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200310
Willy Tarreaua150cf12014-05-20 21:57:23 +0200311 srv_append_status(&trash, s, check_reason_string(check), -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200312
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100313 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200314 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
315 (check->health >= check->rise) ? check->fall : check->rise,
316 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200317
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100318 Warning("%s.\n", trash.str);
319 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.str);
Simon Horman7ea9be02015-04-30 13:10:33 +0900320 send_email_alert(s, LOG_INFO, "%s", trash.str);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200321 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200322}
323
Willy Tarreau4eec5472014-05-20 22:32:27 +0200324/* Marks the check <check>'s server down if the current check is already failed
325 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200326 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200327static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200328{
Simon Horman4a741432013-02-23 15:35:38 +0900329 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900330
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200331 /* The agent secondary check should only cause a server to be marked
332 * as down if check->status is HCHK_STATUS_L7STS, which indicates
333 * that the agent returned "fail", "stopped" or "down".
334 * The implication here is that failure to connect to the agent
335 * as a secondary check should not cause the server to be marked
336 * down. */
337 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
338 return;
339
Willy Tarreau4eec5472014-05-20 22:32:27 +0200340 if (check->health > 0)
341 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100342
Willy Tarreau4eec5472014-05-20 22:32:27 +0200343 /* We only report a reason for the check if we did not do so previously */
344 srv_set_stopped(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200345}
346
Willy Tarreauaf549582014-05-16 17:37:50 +0200347/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200348 * it isn't in maintenance, it is not tracking a down server and other checks
349 * comply. The rule is simple : by default, a server is up, unless any of the
350 * following conditions is true :
351 * - health check failed (check->health < rise)
352 * - agent check failed (agent->health < rise)
353 * - the server tracks a down server (track && track->state == STOPPED)
354 * Note that if the server has a slowstart, it will switch to STARTING instead
355 * of RUNNING. Also, only the health checks support the nolb mode, so the
356 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200357 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200358static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200359{
Simon Horman4a741432013-02-23 15:35:38 +0900360 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100361
Willy Tarreauaf549582014-05-16 17:37:50 +0200362 if (s->admin & SRV_ADMF_MAINT)
363 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100364
Willy Tarreau3e048382014-05-21 10:30:54 +0200365 if (s->track && s->track->state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200366 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100367
Willy Tarreau3e048382014-05-21 10:30:54 +0200368 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
369 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100370
Willy Tarreau3e048382014-05-21 10:30:54 +0200371 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
372 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200373
Willy Tarreau3e048382014-05-21 10:30:54 +0200374 if ((check->state & CHK_ST_AGENT) && s->state == SRV_ST_STOPPING)
375 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100376
Willy Tarreau3e048382014-05-21 10:30:54 +0200377 srv_set_running(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100378}
379
Willy Tarreaudb58b792014-05-21 13:57:23 +0200380/* Marks the check <check> as valid and tries to set its server into stopping mode
381 * if it was running or starting, and provided it isn't in maintenance and other
382 * checks comply. The conditions for the server to be marked in stopping mode are
383 * the same as for it to be turned up. Also, only the health checks support the
384 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200385 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200386static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200387{
Simon Horman4a741432013-02-23 15:35:38 +0900388 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100389
Willy Tarreauaf549582014-05-16 17:37:50 +0200390 if (s->admin & SRV_ADMF_MAINT)
391 return;
392
Willy Tarreaudb58b792014-05-21 13:57:23 +0200393 if (check->state & CHK_ST_AGENT)
394 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100395
Willy Tarreaudb58b792014-05-21 13:57:23 +0200396 if (s->track && s->track->state == SRV_ST_STOPPED)
397 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100398
Willy Tarreaudb58b792014-05-21 13:57:23 +0200399 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
400 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100401
Willy Tarreaudb58b792014-05-21 13:57:23 +0200402 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
403 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100404
Willy Tarreaudb58b792014-05-21 13:57:23 +0200405 srv_set_stopping(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100406}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200407
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100408/* note: use health_adjust() only, which first checks that the observe mode is
409 * enabled.
410 */
411void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100412{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100413 int failed;
414 int expire;
415
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100416 if (s->observe >= HANA_OBS_SIZE)
417 return;
418
Willy Tarreaubb956662013-01-24 00:37:39 +0100419 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100420 return;
421
422 switch (analyze_statuses[status].lr[s->observe - 1]) {
423 case 1:
424 failed = 1;
425 break;
426
427 case 2:
428 failed = 0;
429 break;
430
431 default:
432 return;
433 }
434
435 if (!failed) {
436 /* good: clear consecutive_errors */
437 s->consecutive_errors = 0;
438 return;
439 }
440
441 s->consecutive_errors++;
442
443 if (s->consecutive_errors < s->consecutive_errors_limit)
444 return;
445
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100446 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
447 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100448
449 switch (s->onerror) {
450 case HANA_ONERR_FASTINTER:
451 /* force fastinter - nothing to do here as all modes force it */
452 break;
453
454 case HANA_ONERR_SUDDTH:
455 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900456 if (s->check.health > s->check.rise)
457 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100458
459 /* no break - fall through */
460
461 case HANA_ONERR_FAILCHK:
462 /* simulate a failed health check */
Simon Horman4a741432013-02-23 15:35:38 +0900463 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200464 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100465 break;
466
467 case HANA_ONERR_MARKDWN:
468 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900469 s->check.health = s->check.rise;
Simon Horman4a741432013-02-23 15:35:38 +0900470 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200471 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100472 break;
473
474 default:
475 /* write a warning? */
476 break;
477 }
478
479 s->consecutive_errors = 0;
480 s->counters.failed_hana++;
481
Simon Horman66183002013-02-23 10:16:43 +0900482 if (s->check.fastinter) {
483 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300484 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200485 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300486 /* requeue check task with new expire */
487 task_queue(s->check.task);
488 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100489 }
490}
491
Willy Tarreaua1dab552014-04-14 15:04:54 +0200492static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100493{
494 int sv_state;
495 int ratio;
496 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800497 char addr[46];
498 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100499 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
500 "UP %d/%d", "UP",
501 "NOLB %d/%d", "NOLB",
502 "no check" };
503
504 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
505 hlen += 24;
506
Willy Tarreauff5ae352013-12-11 20:36:34 +0100507 if (!(s->check.state & CHK_ST_ENABLED))
508 sv_state = 6;
Willy Tarreau892337c2014-05-13 23:41:20 +0200509 else if (s->state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900510 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100511 sv_state = 3; /* UP */
512 else
513 sv_state = 2; /* going down */
514
Willy Tarreau892337c2014-05-13 23:41:20 +0200515 if (s->state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100516 sv_state += 2;
517 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900518 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100519 sv_state = 1; /* going up */
520 else
521 sv_state = 0; /* DOWN */
522 }
523
Willy Tarreaua1dab552014-04-14 15:04:54 +0200524 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100525 srv_hlt_st[sv_state],
Willy Tarreau892337c2014-05-13 23:41:20 +0200526 (s->state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
527 (s->state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100528
Joseph Lynch514061c2015-01-15 17:52:59 -0800529 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100530 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
531 snprintf(port, sizeof(port), "%u", s->svc_port);
532 else
533 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800534
535 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
536 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100537 global.node,
538 (s->eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
539 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
540 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
541 s->nbpend);
542
Willy Tarreau892337c2014-05-13 23:41:20 +0200543 if ((s->state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100544 now.tv_sec < s->last_change + s->slowstart &&
545 now.tv_sec >= s->last_change) {
546 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200547 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100548 }
549
550 buffer[hlen++] = '\r';
551 buffer[hlen++] = '\n';
552
553 return hlen;
554}
555
Willy Tarreau20a18342013-12-05 00:31:46 +0100556/* Check the connection. If an error has already been reported or the socket is
557 * closed, keep errno intact as it is supposed to contain the valid error code.
558 * If no error is reported, check the socket's error queue using getsockopt().
559 * Warning, this must be done only once when returning from poll, and never
560 * after an I/O error was attempted, otherwise the error queue might contain
561 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
562 * socket. Returns non-zero if an error was reported, zero if everything is
563 * clean (including a properly closed socket).
564 */
565static int retrieve_errno_from_socket(struct connection *conn)
566{
567 int skerr;
568 socklen_t lskerr = sizeof(skerr);
569
570 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
571 return 1;
572
Willy Tarreau3c728722014-01-23 13:50:42 +0100573 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100574 return 0;
575
576 if (getsockopt(conn->t.sock.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
577 errno = skerr;
578
579 if (errno == EAGAIN)
580 errno = 0;
581
582 if (!errno) {
583 /* we could not retrieve an error, that does not mean there is
584 * none. Just don't change anything and only report the prior
585 * error if any.
586 */
587 if (conn->flags & CO_FL_ERROR)
588 return 1;
589 else
590 return 0;
591 }
592
593 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
594 return 1;
595}
596
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100597/* Try to collect as much information as possible on the connection status,
598 * and adjust the server status accordingly. It may make use of <errno_bck>
599 * if non-null when the caller is absolutely certain of its validity (eg:
600 * checked just after a syscall). If the caller doesn't have a valid errno,
601 * it can pass zero, and retrieve_errno_from_socket() will be called to try
602 * to extract errno from the socket. If no error is reported, it will consider
603 * the <expired> flag. This is intended to be used when a connection error was
604 * reported in conn->flags or when a timeout was reported in <expired>. The
605 * function takes care of not updating a server status which was already set.
606 * All situations where at least one of <expired> or CO_FL_ERROR are set
607 * produce a status.
608 */
609static void chk_report_conn_err(struct connection *conn, int errno_bck, int expired)
610{
611 struct check *check = conn->owner;
612 const char *err_msg;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200613 struct chunk *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200614 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200615 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100616
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100617 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100618 return;
619
620 errno = errno_bck;
621 if (!errno || errno == EAGAIN)
622 retrieve_errno_from_socket(conn);
623
624 if (!(conn->flags & CO_FL_ERROR) && !expired)
625 return;
626
627 /* we'll try to build a meaningful error message depending on the
628 * context of the error possibly present in conn->err_code, and the
629 * socket error possibly collected above. This is useful to know the
630 * exact step of the L6 layer (eg: SSL handshake).
631 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200632 chk = get_trash_chunk();
633
634 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900635 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200636 if (!step)
637 chunk_printf(chk, " at initial connection step of tcp-check");
638 else {
639 chunk_printf(chk, " at step %d of tcp-check", step);
640 /* we were looking for a string */
641 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
642 if (check->last_started_step->port)
643 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
644 else
645 chunk_appendf(chk, " (connect)");
646 }
647 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
648 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200649 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200650 else if (check->last_started_step->expect_regex)
651 chunk_appendf(chk, " (expect regex)");
652 }
653 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
654 chunk_appendf(chk, " (send)");
655 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200656
657 comment = tcpcheck_get_step_comment(check, step);
658 if (comment)
659 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200660 }
661 }
662
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100663 if (conn->err_code) {
664 if (errno && errno != EAGAIN)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200665 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100666 else
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200667 chunk_printf(&trash, "%s%s", conn_err_code_str(conn), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100668 err_msg = trash.str;
669 }
670 else {
671 if (errno && errno != EAGAIN) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200672 chunk_printf(&trash, "%s%s", strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100673 err_msg = trash.str;
674 }
675 else {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200676 err_msg = chk->str;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100677 }
678 }
679
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200680 if (check->state & CHK_ST_PORT_MISS) {
681 /* NOTE: this is reported after <fall> tries */
682 chunk_printf(chk, "No port available for the TCP connection");
683 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
684 }
685
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100686 if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
687 /* L4 not established (yet) */
688 if (conn->flags & CO_FL_ERROR)
689 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
690 else if (expired)
691 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200692
693 /*
694 * might be due to a server IP change.
695 * Let's trigger a DNS resolution if none are currently running.
696 */
697 if ((check->server->resolution) && (check->server->resolution->step == RSLV_STEP_NONE))
698 trigger_resolution(check->server);
699
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100700 }
701 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
702 /* L6 not established (yet) */
703 if (conn->flags & CO_FL_ERROR)
704 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
705 else if (expired)
706 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
707 }
708 else if (conn->flags & CO_FL_ERROR) {
709 /* I/O error after connection was established and before we could diagnose */
710 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
711 }
712 else if (expired) {
713 /* connection established but expired check */
714 if (check->type == PR_O2_SSL3_CHK)
715 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
716 else /* HTTP, SMTP, ... */
717 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
718 }
719
720 return;
721}
722
Willy Tarreaubaaee002006-06-26 02:48:02 +0200723/*
724 * This function is used only for server health-checks. It handles
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200725 * the connection acknowledgement. If the proxy requires L7 health-checks,
726 * it sends the request. In other cases, it calls set_server_check_status()
Simon Horman4a741432013-02-23 15:35:38 +0900727 * to set check->status, check->duration and check->result.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200728 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200729static void event_srv_chk_w(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200730{
Simon Horman4a741432013-02-23 15:35:38 +0900731 struct check *check = conn->owner;
732 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900733 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200734
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100735 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100736 goto out_wakeup;
737
Willy Tarreau310987a2014-01-22 19:46:33 +0100738 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100739 return;
740
Willy Tarreau20a18342013-12-05 00:31:46 +0100741 if (retrieve_errno_from_socket(conn)) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100742 chk_report_conn_err(conn, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100743 __conn_data_stop_both(conn);
744 goto out_wakeup;
745 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100746
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100747 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100748 /* if the output is closed, we can't do anything */
749 conn->flags |= CO_FL_ERROR;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100750 chk_report_conn_err(conn, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100751 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200752 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200753
Willy Tarreau06559ac2013-12-05 01:53:08 +0100754 /* here, we know that the connection is established. That's enough for
755 * a pure TCP check.
756 */
757 if (!check->type)
758 goto out_wakeup;
759
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200760 if (check->type == PR_O2_TCPCHK_CHK) {
761 tcpcheck_main(conn);
762 return;
763 }
764
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100765 if (check->bo->o) {
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100766 conn->xprt->snd_buf(conn, check->bo, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100767 if (conn->flags & CO_FL_ERROR) {
768 chk_report_conn_err(conn, errno, 0);
769 __conn_data_stop_both(conn);
770 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200771 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100772 if (check->bo->o)
773 return;
774 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200775
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100776 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
777 if (s->proxy->timeout.check) {
778 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
779 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200780 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100781 goto out_nowake;
782
Willy Tarreau83749182007-04-15 20:56:27 +0200783 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200784 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200785 out_nowake:
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200786 __conn_data_stop_send(conn); /* nothing more to write */
Willy Tarreaubaaee002006-06-26 02:48:02 +0200787}
788
Willy Tarreaubaaee002006-06-26 02:48:02 +0200789/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200790 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200791 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900792 * set_server_check_status() to update check->status, check->duration
793 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200794
795 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
796 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
797 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
798 * response to an SSL HELLO (the principle is that this is enough to
799 * distinguish between an SSL server and a pure TCP relay). All other cases will
800 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
801 * etc.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200802 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200803static void event_srv_chk_r(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200804{
Simon Horman4a741432013-02-23 15:35:38 +0900805 struct check *check = conn->owner;
806 struct server *s = check->server;
807 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200808 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100809 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200810 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200811
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100812 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200813 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200814
Willy Tarreau310987a2014-01-22 19:46:33 +0100815 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200816 return;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200817
818 if (check->type == PR_O2_TCPCHK_CHK) {
819 tcpcheck_main(conn);
820 return;
821 }
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200822
Willy Tarreau83749182007-04-15 20:56:27 +0200823 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
824 * but the connection was closed on the remote end. Fortunately, recv still
825 * works correctly and we don't need to do the getsockopt() on linux.
826 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000827
828 /* Set buffer to point to the end of the data already read, and check
829 * that there is free space remaining. If the buffer is full, proceed
830 * with running the checks without attempting another socket read.
831 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000832
Willy Tarreau03938182010-03-17 21:52:07 +0100833 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000834
Simon Horman4a741432013-02-23 15:35:38 +0900835 conn->xprt->rcv_buf(conn, check->bi, check->bi->size);
Willy Tarreauf1503172012-09-28 19:39:36 +0200836 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
Willy Tarreau03938182010-03-17 21:52:07 +0100837 done = 1;
Simon Horman4a741432013-02-23 15:35:38 +0900838 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200839 /* Report network errors only if we got no other data. Otherwise
840 * we'll let the upper layers decide whether the response is OK
841 * or not. It is very common that an RST sent by the server is
842 * reported as an error just after the last data chunk.
843 */
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100844 chk_report_conn_err(conn, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100845 goto out_wakeup;
846 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200847 }
848
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100849
Willy Tarreau03938182010-03-17 21:52:07 +0100850 /* Intermediate or complete response received.
Simon Horman4a741432013-02-23 15:35:38 +0900851 * Terminate string in check->bi->data buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100852 */
Simon Horman4a741432013-02-23 15:35:38 +0900853 if (check->bi->i < check->bi->size)
854 check->bi->data[check->bi->i] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100855 else {
Simon Horman4a741432013-02-23 15:35:38 +0900856 check->bi->data[check->bi->i - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100857 done = 1; /* buffer full, don't wait for more data */
858 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200859
Nick Chalk57b1bf72010-03-16 15:50:46 +0000860 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900861 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200862 case PR_O2_HTTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900863 if (!done && check->bi->i < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100864 goto wait_more_data;
865
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100866 /* Check if the server speaks HTTP 1.X */
Simon Horman4a741432013-02-23 15:35:38 +0900867 if ((check->bi->i < strlen("HTTP/1.0 000\r")) ||
868 (memcmp(check->bi->data, "HTTP/1.", 7) != 0 ||
869 (*(check->bi->data + 12) != ' ' && *(check->bi->data + 12) != '\r')) ||
870 !isdigit((unsigned char) *(check->bi->data + 9)) || !isdigit((unsigned char) *(check->bi->data + 10)) ||
871 !isdigit((unsigned char) *(check->bi->data + 11))) {
872 cut_crlf(check->bi->data);
873 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200874
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100875 goto out_wakeup;
876 }
877
Simon Horman4a741432013-02-23 15:35:38 +0900878 check->code = str2uic(check->bi->data + 9);
879 desc = ltrim(check->bi->data + 12, ' ');
Nick Chalk57b1bf72010-03-16 15:50:46 +0000880
Willy Tarreaubd741542010-03-16 18:46:54 +0100881 if ((s->proxy->options & PR_O_DISABLE404) &&
Willy Tarreau892337c2014-05-13 23:41:20 +0200882 (s->state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000883 /* 404 may be accepted as "stopping" only if the server was up */
884 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900885 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000886 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100887 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
888 /* Run content verification check... We know we have at least 13 chars */
889 if (!httpchk_expect(s, done))
890 goto wait_more_data;
891 }
892 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Simon Horman4a741432013-02-23 15:35:38 +0900893 else if (*(check->bi->data + 9) == '2' || *(check->bi->data + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100894 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900895 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100896 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000897 else {
898 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900899 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000900 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200901 break;
902
903 case PR_O2_SSL3_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900904 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100905 goto wait_more_data;
906
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100907 /* Check for SSLv3 alert or handshake */
Simon Horman4a741432013-02-23 15:35:38 +0900908 if ((check->bi->i >= 5) && (*check->bi->data == 0x15 || *check->bi->data == 0x16))
909 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200910 else
Simon Horman4a741432013-02-23 15:35:38 +0900911 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200912 break;
913
914 case PR_O2_SMTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900915 if (!done && check->bi->i < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100916 goto wait_more_data;
917
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200918 /* Check if the server speaks SMTP */
Simon Horman4a741432013-02-23 15:35:38 +0900919 if ((check->bi->i < strlen("000\r")) ||
920 (*(check->bi->data + 3) != ' ' && *(check->bi->data + 3) != '\r') ||
921 !isdigit((unsigned char) *check->bi->data) || !isdigit((unsigned char) *(check->bi->data + 1)) ||
922 !isdigit((unsigned char) *(check->bi->data + 2))) {
923 cut_crlf(check->bi->data);
924 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200925
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200926 goto out_wakeup;
927 }
928
Simon Horman4a741432013-02-23 15:35:38 +0900929 check->code = str2uic(check->bi->data);
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200930
Simon Horman4a741432013-02-23 15:35:38 +0900931 desc = ltrim(check->bi->data + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200932 cut_crlf(desc);
933
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100934 /* Check for SMTP code 2xx (should be 250) */
Simon Horman4a741432013-02-23 15:35:38 +0900935 if (*check->bi->data == '2')
936 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200937 else
Simon Horman4a741432013-02-23 15:35:38 +0900938 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200939 break;
940
Simon Hormana2b9dad2013-02-12 10:45:54 +0900941 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100942 int status = HCHK_STATUS_CHECKED;
943 const char *hs = NULL; /* health status */
944 const char *as = NULL; /* admin status */
945 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200946 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100947 const char *err = NULL; /* first error to report */
948 const char *wrn = NULL; /* first warning to report */
949 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900950
Willy Tarreau81f5d942013-12-09 20:51:51 +0100951 /* We're getting an agent check response. The agent could
952 * have been disabled in the mean time with a long check
953 * still pending. It is important that we ignore the whole
954 * response.
955 */
956 if (!(check->server->agent.state & CHK_ST_ENABLED))
957 break;
958
959 /* The agent supports strings made of a single line ended by the
960 * first CR ('\r') or LF ('\n'). This line is composed of words
961 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
962 * line may optionally contained a description of a state change
963 * after a sharp ('#'), which is only considered if a health state
964 * is announced.
965 *
966 * Words may be composed of :
967 * - a numeric weight suffixed by the percent character ('%').
968 * - a health status among "up", "down", "stopped", and "fail".
969 * - an admin status among "ready", "drain", "maint".
970 *
971 * These words may appear in any order. If multiple words of the
972 * same category appear, the last one wins.
973 */
974
Willy Tarreau9809b782013-12-11 21:40:11 +0100975 p = check->bi->data;
976 while (*p && *p != '\n' && *p != '\r')
977 p++;
978
979 if (!*p) {
980 if (!done)
981 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900982
Willy Tarreau9809b782013-12-11 21:40:11 +0100983 /* at least inform the admin that the agent is mis-behaving */
984 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
985 break;
986 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100987
Willy Tarreau9809b782013-12-11 21:40:11 +0100988 *p = 0;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100989 cmd = check->bi->data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900990
Willy Tarreau81f5d942013-12-09 20:51:51 +0100991 while (*cmd) {
992 /* look for next word */
993 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
994 cmd++;
995 continue;
996 }
Simon Horman671b6f02013-11-25 10:46:39 +0900997
Willy Tarreau81f5d942013-12-09 20:51:51 +0100998 if (*cmd == '#') {
999 /* this is the beginning of a health status description,
1000 * skip the sharp and blanks.
1001 */
1002 cmd++;
1003 while (*cmd == '\t' || *cmd == ' ')
1004 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001005 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001006 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001007
1008 /* find the end of the word so that we have a null-terminated
1009 * word between <cmd> and <p>.
1010 */
1011 p = cmd + 1;
1012 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1013 p++;
1014 if (*p)
1015 *p++ = 0;
1016
1017 /* first, health statuses */
1018 if (strcasecmp(cmd, "up") == 0) {
1019 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001020 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001021 hs = cmd;
1022 }
1023 else if (strcasecmp(cmd, "down") == 0) {
1024 check->health = 0;
1025 status = HCHK_STATUS_L7STS;
1026 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001027 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001028 else if (strcasecmp(cmd, "stopped") == 0) {
1029 check->health = 0;
1030 status = HCHK_STATUS_L7STS;
1031 hs = cmd;
1032 }
1033 else if (strcasecmp(cmd, "fail") == 0) {
1034 check->health = 0;
1035 status = HCHK_STATUS_L7STS;
1036 hs = cmd;
1037 }
1038 /* admin statuses */
1039 else if (strcasecmp(cmd, "ready") == 0) {
1040 as = cmd;
1041 }
1042 else if (strcasecmp(cmd, "drain") == 0) {
1043 as = cmd;
1044 }
1045 else if (strcasecmp(cmd, "maint") == 0) {
1046 as = cmd;
1047 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001048 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001049 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1050 ps = cmd;
1051 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001052 /* try to parse a maxconn here */
1053 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1054 cs = cmd;
1055 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001056 else {
1057 /* keep a copy of the first error */
1058 if (!err)
1059 err = cmd;
1060 }
1061 /* skip to next word */
1062 cmd = p;
1063 }
1064 /* here, cmd points either to \0 or to the beginning of a
1065 * description. Skip possible leading spaces.
1066 */
1067 while (*cmd == ' ' || *cmd == '\n')
1068 cmd++;
1069
1070 /* First, update the admin status so that we avoid sending other
1071 * possibly useless warnings and can also update the health if
1072 * present after going back up.
1073 */
1074 if (as) {
1075 if (strcasecmp(as, "drain") == 0)
1076 srv_adm_set_drain(check->server);
1077 else if (strcasecmp(as, "maint") == 0)
1078 srv_adm_set_maint(check->server);
1079 else
1080 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001081 }
1082
Willy Tarreau81f5d942013-12-09 20:51:51 +01001083 /* now change weights */
1084 if (ps) {
1085 const char *msg;
1086
1087 msg = server_parse_weight_change_request(s, ps);
1088 if (!wrn || !*wrn)
1089 wrn = msg;
1090 }
1091
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001092 if (cs) {
1093 const char *msg;
1094
1095 cs += strlen("maxconn:");
1096
1097 msg = server_parse_maxconn_change_request(s, cs);
1098 if (!wrn || !*wrn)
1099 wrn = msg;
1100 }
1101
Willy Tarreau81f5d942013-12-09 20:51:51 +01001102 /* and finally health status */
1103 if (hs) {
1104 /* We'll report some of the warnings and errors we have
1105 * here. Down reports are critical, we leave them untouched.
1106 * Lack of report, or report of 'UP' leaves the room for
1107 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001108 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001109 const char *msg = cmd;
1110 struct chunk *t;
1111
1112 if (!*msg || status == HCHK_STATUS_L7OKD) {
1113 if (err && *err)
1114 msg = err;
1115 else if (wrn && *wrn)
1116 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001117 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001118
1119 t = get_trash_chunk();
1120 chunk_printf(t, "via agent : %s%s%s%s",
1121 hs, *msg ? " (" : "",
1122 msg, *msg ? ")" : "");
1123
1124 set_server_check_status(check, status, t->str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001125 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001126 else if (err && *err) {
1127 /* No status change but we'd like to report something odd.
1128 * Just report the current state and copy the message.
1129 */
1130 chunk_printf(&trash, "agent reports an error : %s", err);
1131 set_server_check_status(check, status/*check->status*/, trash.str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001132
Willy Tarreau81f5d942013-12-09 20:51:51 +01001133 }
1134 else if (wrn && *wrn) {
1135 /* No status change but we'd like to report something odd.
1136 * Just report the current state and copy the message.
1137 */
1138 chunk_printf(&trash, "agent warns : %s", wrn);
1139 set_server_check_status(check, status/*check->status*/, trash.str);
1140 }
1141 else
1142 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001143 break;
1144 }
1145
Willy Tarreau1620ec32011-08-06 17:05:02 +02001146 case PR_O2_PGSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001147 if (!done && check->bi->i < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001148 goto wait_more_data;
1149
Simon Horman4a741432013-02-23 15:35:38 +09001150 if (check->bi->data[0] == 'R') {
1151 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001152 }
1153 else {
Simon Horman4a741432013-02-23 15:35:38 +09001154 if ((check->bi->data[0] == 'E') && (check->bi->data[5]!=0) && (check->bi->data[6]!=0))
1155 desc = &check->bi->data[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001156 else
1157 desc = "PostgreSQL unknown error";
1158
Simon Horman4a741432013-02-23 15:35:38 +09001159 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001160 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001161 break;
1162
1163 case PR_O2_REDIS_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001164 if (!done && check->bi->i < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001165 goto wait_more_data;
1166
Simon Horman4a741432013-02-23 15:35:38 +09001167 if (strcmp(check->bi->data, "+PONG\r\n") == 0) {
1168 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001169 }
1170 else {
Simon Horman4a741432013-02-23 15:35:38 +09001171 set_server_check_status(check, HCHK_STATUS_L7STS, check->bi->data);
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001172 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001173 break;
1174
1175 case PR_O2_MYSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001176 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001177 goto wait_more_data;
1178
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001179 if (s->proxy->check_len == 0) { // old mode
Simon Horman4a741432013-02-23 15:35:38 +09001180 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001181 /* We set the MySQL Version in description for information purpose
1182 * FIXME : it can be cool to use MySQL Version for other purpose,
1183 * like mark as down old MySQL server.
1184 */
Simon Horman4a741432013-02-23 15:35:38 +09001185 if (check->bi->i > 51) {
1186 desc = ltrim(check->bi->data + 5, ' ');
1187 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001188 }
1189 else {
1190 if (!done)
1191 goto wait_more_data;
1192 /* it seems we have a OK packet but without a valid length,
1193 * it must be a protocol error
1194 */
Simon Horman4a741432013-02-23 15:35:38 +09001195 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001196 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001197 }
1198 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001199 /* An error message is attached in the Error packet */
Simon Horman4a741432013-02-23 15:35:38 +09001200 desc = ltrim(check->bi->data + 7, ' ');
1201 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001202 }
1203 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001204 unsigned int first_packet_len = ((unsigned int) *check->bi->data) +
1205 (((unsigned int) *(check->bi->data + 1)) << 8) +
1206 (((unsigned int) *(check->bi->data + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001207
Simon Horman4a741432013-02-23 15:35:38 +09001208 if (check->bi->i == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001209 /* MySQL Error packet always begin with field_count = 0xff */
Simon Horman4a741432013-02-23 15:35:38 +09001210 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001211 /* We have only one MySQL packet and it is a Handshake Initialization packet
1212 * but we need to have a second packet to know if it is alright
1213 */
Simon Horman4a741432013-02-23 15:35:38 +09001214 if (!done && check->bi->i < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001215 goto wait_more_data;
1216 }
1217 else {
1218 /* We have only one packet and it is an Error packet,
1219 * an error message is attached, so we can display it
1220 */
Simon Horman4a741432013-02-23 15:35:38 +09001221 desc = &check->bi->data[7];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001222 //Warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001223 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001224 }
Simon Horman4a741432013-02-23 15:35:38 +09001225 } else if (check->bi->i > first_packet_len + 4) {
1226 unsigned int second_packet_len = ((unsigned int) *(check->bi->data + first_packet_len + 4)) +
1227 (((unsigned int) *(check->bi->data + first_packet_len + 5)) << 8) +
1228 (((unsigned int) *(check->bi->data + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229
Simon Horman4a741432013-02-23 15:35:38 +09001230 if (check->bi->i == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001231 /* We have 2 packets and that's good */
1232 /* Check if the second packet is a MySQL Error packet or not */
Simon Horman4a741432013-02-23 15:35:38 +09001233 if (*(check->bi->data + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001234 /* No error packet */
1235 /* We set the MySQL Version in description for information purpose */
Simon Horman4a741432013-02-23 15:35:38 +09001236 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001237 //Warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001238 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239 }
1240 else {
1241 /* An error message is attached in the Error packet
1242 * so we can display it ! :)
1243 */
Simon Horman4a741432013-02-23 15:35:38 +09001244 desc = &check->bi->data[first_packet_len+11];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001245 //Warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001246 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001247 }
1248 }
1249 }
1250 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001251 if (!done)
1252 goto wait_more_data;
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001253 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001254 * it must be a protocol error
1255 */
Simon Horman4a741432013-02-23 15:35:38 +09001256 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001257 //Warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001258 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001259 }
1260 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001261 break;
1262
1263 case PR_O2_LDAP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001264 if (!done && check->bi->i < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001265 goto wait_more_data;
1266
1267 /* Check if the server speaks LDAP (ASN.1/BER)
1268 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1269 * http://tools.ietf.org/html/rfc4511
1270 */
1271
1272 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1273 * LDAPMessage: 0x30: SEQUENCE
1274 */
Simon Horman4a741432013-02-23 15:35:38 +09001275 if ((check->bi->i < 14) || (*(check->bi->data) != '\x30')) {
1276 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001277 }
1278 else {
1279 /* size of LDAPMessage */
Simon Horman4a741432013-02-23 15:35:38 +09001280 msglen = (*(check->bi->data + 1) & 0x80) ? (*(check->bi->data + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001281
1282 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1283 * messageID: 0x02 0x01 0x01: INTEGER 1
1284 * protocolOp: 0x61: bindResponse
1285 */
1286 if ((msglen > 2) ||
Simon Horman4a741432013-02-23 15:35:38 +09001287 (memcmp(check->bi->data + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
1288 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001289
1290 goto out_wakeup;
1291 }
1292
1293 /* size of bindResponse */
Simon Horman4a741432013-02-23 15:35:38 +09001294 msglen += (*(check->bi->data + msglen + 6) & 0x80) ? (*(check->bi->data + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001295
1296 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1297 * ldapResult: 0x0a 0x01: ENUMERATION
1298 */
1299 if ((msglen > 4) ||
Simon Horman4a741432013-02-23 15:35:38 +09001300 (memcmp(check->bi->data + 7 + msglen, "\x0a\x01", 2) != 0)) {
1301 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001302
1303 goto out_wakeup;
1304 }
1305
1306 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1307 * resultCode
1308 */
Simon Horman4a741432013-02-23 15:35:38 +09001309 check->code = *(check->bi->data + msglen + 9);
1310 if (check->code) {
1311 set_server_check_status(check, HCHK_STATUS_L7STS, "See RFC: http://tools.ietf.org/html/rfc4511#section-4.1.9");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001312 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001313 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001314 }
1315 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001316 break;
1317
Christopher Fauletba7bc162016-11-07 21:07:38 +01001318 case PR_O2_SPOP_CHK: {
1319 unsigned int framesz;
1320 char err[HCHK_DESC_LEN];
1321
1322 if (!done && check->bi->i < 4)
1323 goto wait_more_data;
1324
1325 memcpy(&framesz, check->bi->data, 4);
1326 framesz = ntohl(framesz);
1327
1328 if (!done && check->bi->i < (4+framesz))
1329 goto wait_more_data;
1330
1331 if (!handle_spoe_healthcheck_response(check->bi->data+4, framesz, err, HCHK_DESC_LEN-1))
1332 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1333 else
1334 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1335 break;
1336 }
1337
Willy Tarreau1620ec32011-08-06 17:05:02 +02001338 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001339 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001340 break;
1341 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001342
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001343 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001344 /* collect possible new errors */
1345 if (conn->flags & CO_FL_ERROR)
1346 chk_report_conn_err(conn, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001347
Nick Chalk57b1bf72010-03-16 15:50:46 +00001348 /* Reset the check buffer... */
Simon Horman4a741432013-02-23 15:35:38 +09001349 *check->bi->data = '\0';
1350 check->bi->i = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +00001351
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001352 /* Close the connection... We absolutely want to perform a hard close
1353 * and reset the connection if some data are pending, otherwise we end
1354 * up with many TIME_WAITs and eat all the source port range quickly.
1355 * To avoid sending RSTs all the time, we first try to drain pending
1356 * data.
1357 */
Willy Tarreaub4017d02015-03-12 23:11:26 +01001358 __conn_data_stop_both(conn);
1359 conn_data_shutw_hard(conn);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001360
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001361 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001362 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001363 conn->flags |= CO_FL_ERROR;
1364
Willy Tarreaufdccded2008-08-29 18:19:04 +02001365 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau3267d362012-08-17 23:53:56 +02001366 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001367
1368 wait_more_data:
Willy Tarreauf817e9f2014-01-10 16:58:45 +01001369 __conn_data_want_recv(conn);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001370}
1371
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001372/*
1373 * This function is used only for server health-checks. It handles connection
1374 * status updates including errors. If necessary, it wakes the check task up.
1375 * It always returns 0.
1376 */
1377static int wake_srv_chk(struct connection *conn)
Willy Tarreau20bea422012-07-06 12:00:49 +02001378{
Simon Horman4a741432013-02-23 15:35:38 +09001379 struct check *check = conn->owner;
Willy Tarreau20bea422012-07-06 12:00:49 +02001380
Willy Tarreau6c560da2012-11-24 11:14:45 +01001381 if (unlikely(conn->flags & CO_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001382 /* We may get error reports bypassing the I/O handlers, typically
1383 * the case when sending a pure TCP check which fails, then the I/O
1384 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001385 * main processing task so let's simply wake it up. If we get here,
1386 * we expect errno to still be valid.
1387 */
1388 chk_report_conn_err(conn, errno, 0);
1389
Willy Tarreau2d351b62013-12-05 02:36:25 +01001390 __conn_data_stop_both(conn);
1391 task_wakeup(check->task, TASK_WOKEN_IO);
1392 }
Willy Tarreau3be293f2014-02-05 18:31:24 +01001393 else if (!(conn->flags & (CO_FL_DATA_RD_ENA|CO_FL_DATA_WR_ENA|CO_FL_HANDSHAKE))) {
1394 /* we may get here if only a connection probe was required : we
1395 * don't have any data to send nor anything expected in response,
1396 * so the completion of the connection establishment is enough.
1397 */
1398 task_wakeup(check->task, TASK_WOKEN_IO);
1399 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001400
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001401 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001402 /* We're here because nobody wants to handle the error, so we
1403 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001404 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001405 conn_sock_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02001406 conn_force_close(conn);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001407 }
Willy Tarreau3267d362012-08-17 23:53:56 +02001408 return 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001409}
1410
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001411struct data_cb check_conn_cb = {
1412 .recv = event_srv_chk_r,
1413 .send = event_srv_chk_w,
1414 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001415 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001416};
1417
Willy Tarreaubaaee002006-06-26 02:48:02 +02001418/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001419 * updates the server's weight during a warmup stage. Once the final weight is
1420 * reached, the task automatically stops. Note that any server status change
1421 * must have updated s->last_change accordingly.
1422 */
1423static struct task *server_warmup(struct task *t)
1424{
1425 struct server *s = t->context;
1426
1427 /* by default, plan on stopping the task */
1428 t->expire = TICK_ETERNITY;
Willy Tarreau20125212014-05-13 19:44:56 +02001429 if ((s->admin & SRV_ADMF_MAINT) ||
Willy Tarreau892337c2014-05-13 23:41:20 +02001430 (s->state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001431 return t;
1432
Willy Tarreau892337c2014-05-13 23:41:20 +02001433 /* recalculate the weights and update the state */
Willy Tarreau004e0452013-11-21 11:22:01 +01001434 server_recalc_eweight(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001435
1436 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001437 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001438
1439 /* get back there in 1 second or 1/20th of the slowstart interval,
1440 * whichever is greater, resulting in small 5% steps.
1441 */
Willy Tarreau892337c2014-05-13 23:41:20 +02001442 if (s->state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001443 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1444 return t;
1445}
1446
1447/*
Simon Horman98637e52014-06-20 12:30:16 +09001448 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001449 *
1450 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001451 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1452 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1453 * - SF_ERR_SRVTO if there are no more servers
1454 * - SF_ERR_SRVCL if the connection was refused by the server
1455 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1456 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1457 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001458 * - SF_ERR_CHK_PORT if no port could be found to run a health check on an AF_INET* socket
Tim Düsterhus4896c442016-11-29 02:15:19 +01001459 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001460 * Note that we try to prevent the network stack from sending the ACK during the
1461 * connect() when a pure TCP check is used (without PROXY protocol).
1462 */
Simon Horman98637e52014-06-20 12:30:16 +09001463static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001464{
1465 struct check *check = t->context;
1466 struct server *s = check->server;
1467 struct connection *conn = check->conn;
1468 struct protocol *proto;
1469 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001470 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001471
1472 /* tcpcheck send/expect initialisation */
1473 if (check->type == PR_O2_TCPCHK_CHK)
1474 check->current_step = NULL;
1475
1476 /* prepare the check buffer.
1477 * This should not be used if check is the secondary agent check
1478 * of a server as s->proxy->check_req will relate to the
1479 * configuration of the primary check. Similarly, tcp-check uses
1480 * its own strings.
1481 */
1482 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
1483 bo_putblk(check->bo, s->proxy->check_req, s->proxy->check_len);
1484
1485 /* we want to check if this host replies to HTTP or SSLv3 requests
1486 * so we'll send the request, and won't wake the checker up now.
1487 */
1488 if ((check->type) == PR_O2_SSL3_CHK) {
1489 /* SSL requires that we put Unix time in the request */
1490 int gmt_time = htonl(date.tv_sec);
1491 memcpy(check->bo->data + 11, &gmt_time, 4);
1492 }
1493 else if ((check->type) == PR_O2_HTTP_CHK) {
1494 if (s->proxy->options2 & PR_O2_CHK_SNDST)
1495 bo_putblk(check->bo, trash.str, httpchk_build_status_header(s, trash.str, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001496 /* prevent HTTP keep-alive when "http-check expect" is used */
1497 if (s->proxy->options2 & PR_O2_EXP_TYPE)
1498 bo_putstr(check->bo, "Connection: close\r\n");
Simon Hormanb00d17a2014-06-13 16:18:16 +09001499 bo_putstr(check->bo, "\r\n");
1500 *check->bo->p = '\0'; /* to make gdb output easier to read */
1501 }
1502 }
1503
James Brown55f9ff12015-10-21 18:19:05 -07001504 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
1505 bo_putblk(check->bo, check->send_string, check->send_string_len);
1506 }
1507
Simon Hormanb00d17a2014-06-13 16:18:16 +09001508 /* prepare a new connection */
1509 conn_init(conn);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001510
Simon Horman41f58762015-01-30 11:22:56 +09001511 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001512 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001513 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001514 }
1515 else {
1516 /* we'll connect to the addr on the server */
1517 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001518 }
1519
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001520 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
1521 int i = 0;
1522
1523 i = srv_check_healthcheck_port(check);
1524 if (i == 0) {
1525 conn->owner = check;
1526 return SF_ERR_CHK_PORT;
1527 }
1528
1529 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001530 }
1531
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001532 proto = protocol_by_family(conn->addr.to.ss_family);
1533
1534 conn_prepare(conn, proto, check->xprt);
1535 conn_attach(conn, check, &check_conn_cb);
1536 conn->target = &s->obj_type;
1537
1538 /* no client address */
1539 clear_addr(&conn->addr.from);
1540
Willy Tarreauf3d34822014-12-08 12:11:28 +01001541 /* only plain tcp-check supports quick ACK */
1542 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1543
Simon Hormane16c1b32015-01-30 11:22:57 +09001544 if (check->type == PR_O2_TCPCHK_CHK && !LIST_ISEMPTY(check->tcpcheck_rules)) {
Willy Tarreau5581c272015-05-13 12:24:53 +02001545 struct tcpcheck_rule *r;
1546
1547 r = LIST_NEXT(check->tcpcheck_rules, struct tcpcheck_rule *, list);
1548
Simon Hormanb00d17a2014-06-13 16:18:16 +09001549 /* if first step is a 'connect', then tcpcheck_main must run it */
1550 if (r->action == TCPCHK_ACT_CONNECT) {
1551 tcpcheck_main(conn);
Willy Tarreaue7dff022015-04-03 01:14:29 +02001552 return SF_ERR_UP;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001553 }
Willy Tarreauf3d34822014-12-08 12:11:28 +01001554 if (r->action == TCPCHK_ACT_EXPECT)
1555 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001556 }
1557
Willy Tarreaue7dff022015-04-03 01:14:29 +02001558 ret = SF_ERR_INTERNAL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001559 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001560 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001561 conn->flags |= CO_FL_WAKE_DATA;
1562 if (s->check.send_proxy) {
1563 conn->send_proxy_ofs = 1;
1564 conn->flags |= CO_FL_SEND_PROXY;
1565 }
1566
1567 return ret;
1568}
1569
Simon Horman98637e52014-06-20 12:30:16 +09001570static struct list pid_list = LIST_HEAD_INIT(pid_list);
1571static struct pool_head *pool2_pid_list;
1572
1573void block_sigchld(void)
1574{
1575 sigset_t set;
1576 sigemptyset(&set);
1577 sigaddset(&set, SIGCHLD);
Willy Tarreauebc92442016-06-21 17:29:46 +02001578 assert(sigprocmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001579}
1580
1581void unblock_sigchld(void)
1582{
1583 sigset_t set;
1584 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001585 sigaddset(&set, SIGCHLD);
1586 assert(sigprocmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001587}
1588
Simon Horman98637e52014-06-20 12:30:16 +09001589static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1590{
1591 struct pid_list *elem;
1592 struct check *check = t->context;
1593
1594 elem = pool_alloc2(pool2_pid_list);
1595 if (!elem)
1596 return NULL;
1597 elem->pid = pid;
1598 elem->t = t;
1599 elem->exited = 0;
1600 check->curpid = elem;
1601 LIST_INIT(&elem->list);
1602 LIST_ADD(&pid_list, &elem->list);
1603 return elem;
1604}
1605
Simon Horman98637e52014-06-20 12:30:16 +09001606static void pid_list_del(struct pid_list *elem)
1607{
1608 struct check *check;
1609
1610 if (!elem)
1611 return;
1612
Simon Horman98637e52014-06-20 12:30:16 +09001613 LIST_DEL(&elem->list);
Simon Horman98637e52014-06-20 12:30:16 +09001614 if (!elem->exited)
1615 kill(elem->pid, SIGTERM);
1616
1617 check = elem->t->context;
1618 check->curpid = NULL;
1619 pool_free2(pool2_pid_list, elem);
1620}
1621
1622/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1623static void pid_list_expire(pid_t pid, int status)
1624{
1625 struct pid_list *elem;
1626
1627 list_for_each_entry(elem, &pid_list, list) {
1628 if (elem->pid == pid) {
1629 elem->t->expire = now_ms;
1630 elem->status = status;
1631 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001632 task_wakeup(elem->t, TASK_WOKEN_IO);
Simon Horman98637e52014-06-20 12:30:16 +09001633 return;
1634 }
1635 }
1636}
1637
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001638static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001639{
1640 pid_t pid;
1641 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001642
Simon Horman98637e52014-06-20 12:30:16 +09001643 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1644 pid_list_expire(pid, status);
1645}
1646
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001647static int init_pid_list(void)
1648{
Simon Horman98637e52014-06-20 12:30:16 +09001649 if (pool2_pid_list != NULL)
1650 /* Nothing to do */
1651 return 0;
1652
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001653 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Simon Horman98637e52014-06-20 12:30:16 +09001654 Alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1655 strerror(errno));
1656 return 1;
1657 }
1658
1659 pool2_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1660 if (pool2_pid_list == NULL) {
1661 Alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1662 strerror(errno));
1663 return 1;
1664 }
1665
1666 return 0;
1667}
1668
Cyril Bontéac92a062014-12-27 22:28:38 +01001669/* helper macro to set an environment variable and jump to a specific label on failure. */
1670#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001671
1672/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001673 * helper function to allocate enough memory to store an environment variable.
1674 * It will also check that the environment variable is updatable, and silently
1675 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001676 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001677static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001678{
1679 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001680 char *envname;
1681 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001682
Cyril Bontéac92a062014-12-27 22:28:38 +01001683 if (idx < 0 || idx >= EXTCHK_SIZE) {
1684 Alert("Illegal environment variable index %d. Aborting.\n", idx);
1685 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001686 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001687
1688 envname = extcheck_envs[idx].name;
1689 vmaxlen = extcheck_envs[idx].vmaxlen;
1690
1691 /* Check if the environment variable is already set, and silently reject
1692 * the update if this one is not updatable. */
1693 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1694 return 0;
1695
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001696 /* Instead of sending NOT_USED, sending an empty value is preferable */
1697 if (strcmp(value, "NOT_USED") == 0) {
1698 value = "";
1699 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001700
1701 len = strlen(envname) + 1;
1702 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1703 len += strlen(value);
1704 else
1705 len += vmaxlen;
1706
1707 if (!check->envp[idx])
1708 check->envp[idx] = malloc(len + 1);
1709
1710 if (!check->envp[idx]) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001711 Alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
1712 return 1;
1713 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001714 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001715 if (ret < 0) {
1716 Alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
1717 return 1;
1718 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001719 else if (ret > len) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001720 Alert("Environment variable '%s' was truncated. Aborting.\n", envname);
1721 return 1;
1722 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001723 return 0;
1724}
Simon Horman98637e52014-06-20 12:30:16 +09001725
1726static int prepare_external_check(struct check *check)
1727{
1728 struct server *s = check->server;
1729 struct proxy *px = s->proxy;
1730 struct listener *listener = NULL, *l;
1731 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001732 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001733 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001734
1735 list_for_each_entry(l, &px->conf.listeners, by_fe)
1736 /* Use the first INET, INET6 or UNIX listener */
1737 if (l->addr.ss_family == AF_INET ||
1738 l->addr.ss_family == AF_INET6 ||
1739 l->addr.ss_family == AF_UNIX) {
1740 listener = l;
1741 break;
1742 }
1743
Simon Horman98637e52014-06-20 12:30:16 +09001744 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001745 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1746 if (!check->envp) {
1747 Alert("Failed to allocate memory for environment variables. Aborting\n");
1748 goto err;
1749 }
Simon Horman98637e52014-06-20 12:30:16 +09001750
Cyril Bontéac92a062014-12-27 22:28:38 +01001751 check->argv = calloc(6, sizeof(char *));
1752 if (!check->argv) {
1753 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001754 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001755 }
Simon Horman98637e52014-06-20 12:30:16 +09001756
1757 check->argv[0] = px->check_command;
1758
Cyril Bonté777be862014-12-02 21:21:35 +01001759 if (!listener) {
1760 check->argv[1] = strdup("NOT_USED");
1761 check->argv[2] = strdup("NOT_USED");
1762 }
1763 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001764 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001765 addr_to_str(&listener->addr, buf, sizeof(buf));
1766 check->argv[1] = strdup(buf);
1767 port_to_str(&listener->addr, buf, sizeof(buf));
1768 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001769 }
1770 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001771 const struct sockaddr_un *un;
1772
1773 un = (struct sockaddr_un *)&listener->addr;
1774 check->argv[1] = strdup(un->sun_path);
1775 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001776 }
1777 else {
Cyril Bontéac92a062014-12-27 22:28:38 +01001778 Alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001779 goto err;
1780 }
1781
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001782 addr_to_str(&s->addr, buf, sizeof(buf));
1783 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001784
1785 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1786 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1787 else
1788 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001789 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001790
Cyril Bontéac92a062014-12-27 22:28:38 +01001791 for (i = 0; i < 5; i++) {
1792 if (!check->argv[i]) {
1793 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001794 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001795 }
1796 }
Simon Horman98637e52014-06-20 12:30:16 +09001797
Cyril Bontéac92a062014-12-27 22:28:38 +01001798 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001799 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001800 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1801 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1802 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1803 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001804 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001805 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1806 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1807 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1808 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1809 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1810 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1811
1812 /* Ensure that we don't leave any hole in check->envp */
1813 for (i = 0; i < EXTCHK_SIZE; i++)
1814 if (!check->envp[i])
1815 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001816
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001817 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001818err:
1819 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001820 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001821 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001822 free(check->envp);
1823 check->envp = NULL;
1824 }
1825
1826 if (check->argv) {
1827 for (i = 1; i < 5; i++)
1828 free(check->argv[i]);
1829 free(check->argv);
1830 check->argv = NULL;
1831 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001832 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001833}
1834
Simon Hormanb00d17a2014-06-13 16:18:16 +09001835/*
Simon Horman98637e52014-06-20 12:30:16 +09001836 * establish a server health-check that makes use of a process.
1837 *
1838 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001839 * - SF_ERR_NONE if everything's OK
1840 * - SF_ERR_SRVTO if there are no more servers
1841 * - SF_ERR_SRVCL if the connection was refused by the server
1842 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1843 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1844 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01001845 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001846 *
1847 * Blocks and then unblocks SIGCHLD
1848 */
1849static int connect_proc_chk(struct task *t)
1850{
Cyril Bontéac92a062014-12-27 22:28:38 +01001851 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001852 struct check *check = t->context;
1853 struct server *s = check->server;
1854 struct proxy *px = s->proxy;
1855 int status;
1856 pid_t pid;
1857
Willy Tarreaue7dff022015-04-03 01:14:29 +02001858 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001859
1860 block_sigchld();
1861
1862 pid = fork();
1863 if (pid < 0) {
1864 Alert("Failed to fork process for external health check: %s. Aborting.\n",
1865 strerror(errno));
1866 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1867 goto out;
1868 }
1869 if (pid == 0) {
1870 /* Child */
1871 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001872 int fd;
1873
1874 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1875 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1876
1877 while (fd < global.rlimit_nofile)
1878 close(fd++);
1879
Simon Horman98637e52014-06-20 12:30:16 +09001880 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001881 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001882 execvp(px->check_command, check->argv);
1883 Alert("Failed to exec process for external health check: %s. Aborting.\n",
1884 strerror(errno));
1885 exit(-1);
1886 }
1887
1888 /* Parent */
1889 if (check->result == CHK_RES_UNKNOWN) {
1890 if (pid_list_add(pid, t) != NULL) {
1891 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1892
1893 if (px->timeout.check && px->timeout.connect) {
1894 int t_con = tick_add(now_ms, px->timeout.connect);
1895 t->expire = tick_first(t->expire, t_con);
1896 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001897 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001898 goto out;
1899 }
1900 else {
1901 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1902 }
1903 kill(pid, SIGTERM); /* process creation error */
1904 }
1905 else
1906 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1907
1908out:
1909 unblock_sigchld();
1910 return status;
1911}
1912
1913/*
Simon Horman98637e52014-06-20 12:30:16 +09001914 * manages a server health-check that uses a process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001915 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
1916 */
Simon Horman98637e52014-06-20 12:30:16 +09001917static struct task *process_chk_proc(struct task *t)
1918{
1919 struct check *check = t->context;
1920 struct server *s = check->server;
1921 struct connection *conn = check->conn;
1922 int rv;
1923 int ret;
1924 int expired = tick_is_expired(t->expire, now_ms);
1925
1926 if (!(check->state & CHK_ST_INPROGRESS)) {
1927 /* no check currently running */
1928 if (!expired) /* woke up too early */
1929 return t;
1930
1931 /* we don't send any health-checks when the proxy is
1932 * stopped, the server should not be checked or the check
1933 * is disabled.
1934 */
1935 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
1936 s->proxy->state == PR_STSTOPPED)
1937 goto reschedule;
1938
1939 /* we'll initiate a new check */
1940 set_server_check_status(check, HCHK_STATUS_START, NULL);
1941
1942 check->state |= CHK_ST_INPROGRESS;
1943
Simon Hormandbf70192015-01-30 11:22:53 +09001944 ret = connect_proc_chk(t);
Simon Horman98637e52014-06-20 12:30:16 +09001945 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02001946 case SF_ERR_UP:
Simon Horman98637e52014-06-20 12:30:16 +09001947 return t;
Willy Tarreaue7dff022015-04-03 01:14:29 +02001948 case SF_ERR_NONE:
Simon Horman98637e52014-06-20 12:30:16 +09001949 /* we allow up to min(inter, timeout.connect) for a connection
1950 * to establish but only when timeout.check is set
1951 * as it may be to short for a full check otherwise
1952 */
1953 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1954
1955 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
1956 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
1957 t->expire = tick_first(t->expire, t_con);
1958 }
1959
1960 goto reschedule;
1961
Willy Tarreaue7dff022015-04-03 01:14:29 +02001962 case SF_ERR_SRVTO: /* ETIMEDOUT */
1963 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Simon Horman98637e52014-06-20 12:30:16 +09001964 conn->flags |= CO_FL_ERROR;
1965 chk_report_conn_err(conn, errno, 0);
1966 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02001967 case SF_ERR_PRXCOND:
1968 case SF_ERR_RESOURCE:
1969 case SF_ERR_INTERNAL:
Simon Horman98637e52014-06-20 12:30:16 +09001970 conn->flags |= CO_FL_ERROR;
1971 chk_report_conn_err(conn, 0, 0);
1972 break;
1973 }
1974
1975 /* here, we have seen a synchronous error, no fd was allocated */
1976
1977 check->state &= ~CHK_ST_INPROGRESS;
1978 check_notify_failure(check);
1979
1980 /* we allow up to min(inter, timeout.connect) for a connection
1981 * to establish but only when timeout.check is set
1982 * as it may be to short for a full check otherwise
1983 */
1984 while (tick_is_expired(t->expire, now_ms)) {
1985 int t_con;
1986
1987 t_con = tick_add(t->expire, s->proxy->timeout.connect);
1988 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
1989
1990 if (s->proxy->timeout.check)
1991 t->expire = tick_first(t->expire, t_con);
1992 }
1993 }
1994 else {
1995 /* there was a test running.
1996 * First, let's check whether there was an uncaught error,
1997 * which can happen on connect timeout or error.
1998 */
1999 if (check->result == CHK_RES_UNKNOWN) {
2000 /* good connection is enough for pure TCP check */
2001 struct pid_list *elem = check->curpid;
2002 int status = HCHK_STATUS_UNKNOWN;
2003
2004 if (elem->exited) {
2005 status = elem->status; /* Save in case the process exits between use below */
2006 if (!WIFEXITED(status))
2007 check->code = -1;
2008 else
2009 check->code = WEXITSTATUS(status);
2010 if (!WIFEXITED(status) || WEXITSTATUS(status))
2011 status = HCHK_STATUS_PROCERR;
2012 else
2013 status = HCHK_STATUS_PROCOK;
2014 } else if (expired) {
2015 status = HCHK_STATUS_PROCTOUT;
Willy Tarreaudc3d1902014-07-08 00:56:27 +02002016 Warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002017 kill(elem->pid, SIGTERM);
2018 }
2019 set_server_check_status(check, status, NULL);
2020 }
2021
2022 if (check->result == CHK_RES_FAILED) {
2023 /* a failure or timeout detected */
2024 check_notify_failure(check);
2025 }
2026 else if (check->result == CHK_RES_CONDPASS) {
2027 /* check is OK but asks for stopping mode */
2028 check_notify_stopping(check);
2029 }
2030 else if (check->result == CHK_RES_PASSED) {
2031 /* a success was detected */
2032 check_notify_success(check);
2033 }
2034 check->state &= ~CHK_ST_INPROGRESS;
2035
2036 pid_list_del(check->curpid);
2037
2038 rv = 0;
2039 if (global.spread_checks > 0) {
2040 rv = srv_getinter(check) * global.spread_checks / 100;
2041 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2042 }
2043 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2044 }
2045
2046 reschedule:
2047 while (tick_is_expired(t->expire, now_ms))
2048 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2049 return t;
2050}
2051
2052/*
2053 * manages a server health-check that uses a connection. Returns
2054 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2055 */
2056static struct task *process_chk_conn(struct task *t)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002057{
Simon Horman4a741432013-02-23 15:35:38 +09002058 struct check *check = t->context;
2059 struct server *s = check->server;
2060 struct connection *conn = check->conn;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002061 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002062 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002063 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002064
Willy Tarreau2c115e52013-12-11 19:41:16 +01002065 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002066 /* no check currently running */
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002067 if (!expired) /* woke up too early */
Willy Tarreau26c25062009-03-08 09:38:41 +01002068 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002069
Simon Horman671b6f02013-11-25 10:46:39 +09002070 /* we don't send any health-checks when the proxy is
2071 * stopped, the server should not be checked or the check
2072 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002073 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002074 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002075 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002076 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002077
2078 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002079 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002080
Willy Tarreau2c115e52013-12-11 19:41:16 +01002081 check->state |= CHK_ST_INPROGRESS;
Simon Horman4a741432013-02-23 15:35:38 +09002082 check->bi->p = check->bi->data;
2083 check->bi->i = 0;
2084 check->bo->p = check->bo->data;
2085 check->bo->o = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002086
Simon Hormandbf70192015-01-30 11:22:53 +09002087 ret = connect_conn_chk(t);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002088 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002089 case SF_ERR_UP:
Simon Hormanb00d17a2014-06-13 16:18:16 +09002090 return t;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002091 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002092 /* we allow up to min(inter, timeout.connect) for a connection
2093 * to establish but only when timeout.check is set
2094 * as it may be to short for a full check otherwise
2095 */
Simon Horman4a741432013-02-23 15:35:38 +09002096 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002097
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002098 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2099 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2100 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002101 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002102
2103 if (check->type)
2104 conn_data_want_recv(conn); /* prepare for reading a possible reply */
2105
Willy Tarreau5a78f362012-11-23 12:47:05 +01002106 goto reschedule;
2107
Willy Tarreaue7dff022015-04-03 01:14:29 +02002108 case SF_ERR_SRVTO: /* ETIMEDOUT */
2109 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau4bd07de2014-01-24 16:10:57 +01002110 conn->flags |= CO_FL_ERROR;
2111 chk_report_conn_err(conn, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002112 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002113 /* should share same code than cases below */
2114 case SF_ERR_CHK_PORT:
2115 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002116 case SF_ERR_PRXCOND:
2117 case SF_ERR_RESOURCE:
2118 case SF_ERR_INTERNAL:
Willy Tarreau4bd07de2014-01-24 16:10:57 +01002119 conn->flags |= CO_FL_ERROR;
2120 chk_report_conn_err(conn, 0, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002121 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002122 }
2123
Willy Tarreau5a78f362012-11-23 12:47:05 +01002124 /* here, we have seen a synchronous error, no fd was allocated */
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002125
Willy Tarreau2c115e52013-12-11 19:41:16 +01002126 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002127 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002128
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002129 /* we allow up to min(inter, timeout.connect) for a connection
2130 * to establish but only when timeout.check is set
2131 * as it may be to short for a full check otherwise
2132 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002133 while (tick_is_expired(t->expire, now_ms)) {
2134 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002135
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002136 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002137 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002138
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002139 if (s->proxy->timeout.check)
2140 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002141 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002142 }
2143 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002144 /* there was a test running.
2145 * First, let's check whether there was an uncaught error,
2146 * which can happen on connect timeout or error.
2147 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002148 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002149 /* good connection is enough for pure TCP check */
2150 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002151 if (check->use_ssl)
2152 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002153 else
Simon Horman4a741432013-02-23 15:35:38 +09002154 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002155 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002156 else if ((conn->flags & CO_FL_ERROR) || expired) {
2157 chk_report_conn_err(conn, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002158 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002159 else
2160 goto out_wait; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002161 }
2162
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002163 /* check complete or aborted */
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002164 if (conn->xprt) {
2165 /* The check was aborted and the connection was not yet closed.
2166 * This can happen upon timeout, or when an external event such
2167 * as a failed response coupled with "observe layer7" caused the
2168 * server state to be suddenly changed.
2169 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002170 conn_sock_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02002171 conn_force_close(conn);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002172 }
2173
Willy Tarreauaf549582014-05-16 17:37:50 +02002174 if (check->result == CHK_RES_FAILED) {
2175 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002176 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002177 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002178 else if (check->result == CHK_RES_CONDPASS) {
2179 /* check is OK but asks for stopping mode */
2180 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002181 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002182 else if (check->result == CHK_RES_PASSED) {
2183 /* a success was detected */
2184 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002185 }
Willy Tarreau2c115e52013-12-11 19:41:16 +01002186 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002187
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002188 rv = 0;
2189 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002190 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002191 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002192 }
Simon Horman4a741432013-02-23 15:35:38 +09002193 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002194 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002195
2196 reschedule:
2197 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002198 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002199 out_wait:
Willy Tarreau26c25062009-03-08 09:38:41 +01002200 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002201}
2202
Simon Horman98637e52014-06-20 12:30:16 +09002203/*
2204 * manages a server health-check. Returns
2205 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2206 */
2207static struct task *process_chk(struct task *t)
2208{
2209 struct check *check = t->context;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002210 struct server *s = check->server;
2211 struct dns_resolution *resolution = s->resolution;
2212
2213 /* trigger name resolution */
2214 if ((s->check.state & CHK_ST_ENABLED) && (resolution)) {
2215 /* check if a no resolution is running for this server */
2216 if (resolution->step == RSLV_STEP_NONE) {
2217 /*
2218 * if there has not been any name resolution for a longer period than
2219 * hold.valid, let's trigger a new one.
2220 */
Baptiste Assmannf0d93702015-09-03 10:59:39 +02002221 if (!resolution->last_resolution || tick_is_expired(tick_add(resolution->last_resolution, resolution->resolvers->hold.valid), now_ms)) {
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002222 trigger_resolution(s);
2223 }
2224 }
2225 }
Simon Horman98637e52014-06-20 12:30:16 +09002226
2227 if (check->type == PR_O2_EXT_CHK)
2228 return process_chk_proc(t);
2229 return process_chk_conn(t);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002230
2231}
2232
2233/*
2234 * Initiates a new name resolution:
2235 * - generates a query id
2236 * - configure the resolution structure
2237 * - startup the resolvers task if required
2238 *
2239 * returns:
2240 * - 0 in case of error or if resolution already running
2241 * - 1 if everything started properly
2242 */
2243int trigger_resolution(struct server *s)
2244{
2245 struct dns_resolution *resolution;
2246 struct dns_resolvers *resolvers;
2247 int query_id;
2248 int i;
2249
2250 resolution = s->resolution;
2251 resolvers = resolution->resolvers;
2252
2253 /*
2254 * check if a resolution has already been started for this server
2255 * return directly to avoid resolution pill up
2256 */
2257 if (resolution->step != RSLV_STEP_NONE)
2258 return 0;
2259
2260 /* generates a query id */
2261 i = 0;
2262 do {
2263 query_id = dns_rnd16();
2264 /* we do try only 100 times to find a free query id */
2265 if (i++ > 100) {
2266 chunk_printf(&trash, "could not generate a query id for %s/%s, in resolvers %s",
2267 s->proxy->id, s->id, resolvers->id);
2268
2269 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.str);
2270 return 0;
2271 }
2272 } while (eb32_lookup(&resolvers->query_ids, query_id));
2273
2274 LIST_ADDQ(&resolvers->curr_resolution, &resolution->list);
2275
2276 /* now update resolution parameters */
2277 resolution->query_id = query_id;
2278 resolution->qid.key = query_id;
2279 resolution->step = RSLV_STEP_RUNNING;
Thierry Fournierada34842016-02-17 21:25:09 +01002280 resolution->opts = &s->dns_opts;
2281 if (resolution->opts->family_prio == AF_INET) {
Andrew Hayworthe6a4a322015-10-19 22:29:51 +00002282 resolution->query_type = DNS_RTYPE_A;
2283 } else {
2284 resolution->query_type = DNS_RTYPE_AAAA;
2285 }
Baptiste Assmannf778bb42015-09-09 00:54:38 +02002286 resolution->try = resolvers->resolve_retries;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002287 resolution->try_cname = 0;
2288 resolution->nb_responses = 0;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002289 eb32_insert(&resolvers->query_ids, &resolution->qid);
2290
2291 dns_send_query(resolution);
Baptiste Assmannf778bb42015-09-09 00:54:38 +02002292 resolution->try -= 1;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002293
2294 /* update wakeup date if this resolution is the only one in the FIFO list */
2295 if (dns_check_resolution_queue(resolvers) == 1) {
2296 /* update task timeout */
2297 dns_update_resolvers_timeout(resolvers);
2298 task_queue(resolvers->t);
2299 }
2300
2301 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09002302}
2303
Simon Horman5c942422013-11-25 10:46:32 +09002304static int start_check_task(struct check *check, int mininter,
2305 int nbcheck, int srvpos)
2306{
2307 struct task *t;
2308 /* task for the check */
2309 if ((t = task_new()) == NULL) {
2310 Alert("Starting [%s:%s] check: out of memory.\n",
2311 check->server->proxy->id, check->server->id);
2312 return 0;
2313 }
2314
2315 check->task = t;
2316 t->process = process_chk;
2317 t->context = check;
2318
Willy Tarreau1746eec2014-04-25 10:46:47 +02002319 if (mininter < srv_getinter(check))
2320 mininter = srv_getinter(check);
2321
2322 if (global.max_spread_checks && mininter > global.max_spread_checks)
2323 mininter = global.max_spread_checks;
2324
Simon Horman5c942422013-11-25 10:46:32 +09002325 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002326 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002327 check->start = now;
2328 task_queue(t);
2329
2330 return 1;
2331}
2332
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002333/*
2334 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002335 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002336 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002337static int start_checks()
2338{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002339
2340 struct proxy *px;
2341 struct server *s;
2342 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002343 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002344
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002345 /* 1- count the checkers to run simultaneously.
2346 * We also determine the minimum interval among all of those which
2347 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2348 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002349 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002350 * too short an interval for all others.
2351 */
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002352 for (px = proxy; px; px = px->next) {
2353 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002354 if (s->slowstart) {
2355 if ((t = task_new()) == NULL) {
2356 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002357 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002358 }
2359 /* We need a warmup task that will be called when the server
2360 * state switches from down to up.
2361 */
2362 s->warmup = t;
2363 t->process = server_warmup;
2364 t->context = s;
2365 t->expire = TICK_ETERNITY;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002366 /* server can be in this state only because of */
2367 if (s->state == SRV_ST_STARTING)
2368 task_schedule(s->warmup, tick_add(now_ms, MS_TO_TICKS(MAX(1000, (now.tv_sec - s->last_change)) / 20)));
Willy Tarreaue7b73482013-11-21 11:50:50 +01002369 }
2370
Willy Tarreaud8514a22013-12-11 21:10:14 +01002371 if (s->check.state & CHK_ST_CONFIGURED) {
2372 nbcheck++;
2373 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2374 (!mininter || mininter > srv_getinter(&s->check)))
2375 mininter = srv_getinter(&s->check);
2376 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002377
Willy Tarreaud8514a22013-12-11 21:10:14 +01002378 if (s->agent.state & CHK_ST_CONFIGURED) {
2379 nbcheck++;
2380 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2381 (!mininter || mininter > srv_getinter(&s->agent)))
2382 mininter = srv_getinter(&s->agent);
2383 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002384 }
2385 }
2386
Simon Horman4a741432013-02-23 15:35:38 +09002387 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002388 return 0;
2389
2390 srand((unsigned)time(NULL));
2391
2392 /*
2393 * 2- start them as far as possible from each others. For this, we will
2394 * start them after their interval set to the min interval divided by
2395 * the number of servers, weighted by the server's position in the list.
2396 */
2397 for (px = proxy; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002398 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2399 if (init_pid_list()) {
2400 Alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002401 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002402 }
2403 }
2404
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002405 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002406 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002407 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002408 if (s->check.type == PR_O2_EXT_CHK) {
2409 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002410 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002411 }
Simon Hormand60d6912013-11-25 10:46:36 +09002412 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002413 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002414 srvpos++;
2415 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002416
Simon Hormand60d6912013-11-25 10:46:36 +09002417 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002418 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002419 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002420 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002421 }
2422 srvpos++;
2423 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002424 }
2425 }
2426 return 0;
2427}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002428
2429/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002430 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002431 * The buffer MUST be terminated by a null byte before calling this function.
2432 * Sets server status appropriately. The caller is responsible for ensuring
2433 * that the buffer contains at least 13 characters. If <done> is zero, we may
2434 * return 0 to indicate that data is required to decide of a match.
2435 */
2436static int httpchk_expect(struct server *s, int done)
2437{
2438 static char status_msg[] = "HTTP status check returned code <000>";
2439 char status_code[] = "000";
2440 char *contentptr;
2441 int crlf;
2442 int ret;
2443
2444 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2445 case PR_O2_EXP_STS:
2446 case PR_O2_EXP_RSTS:
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002447 memcpy(status_code, s->check.bi->data + 9, 3);
2448 memcpy(status_msg + strlen(status_msg) - 4, s->check.bi->data + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002449
2450 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2451 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2452 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002453 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002454
2455 /* we necessarily have the response, so there are no partial failures */
2456 if (s->proxy->options2 & PR_O2_EXP_INV)
2457 ret = !ret;
2458
Simon Horman4a741432013-02-23 15:35:38 +09002459 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002460 break;
2461
2462 case PR_O2_EXP_STR:
2463 case PR_O2_EXP_RSTR:
2464 /* very simple response parser: ignore CR and only count consecutive LFs,
2465 * stop with contentptr pointing to first char after the double CRLF or
2466 * to '\0' if crlf < 2.
2467 */
2468 crlf = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002469 for (contentptr = s->check.bi->data; *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002470 if (crlf >= 2)
2471 break;
2472 if (*contentptr == '\r')
2473 continue;
2474 else if (*contentptr == '\n')
2475 crlf++;
2476 else
2477 crlf = 0;
2478 }
2479
2480 /* Check that response contains a body... */
2481 if (crlf < 2) {
2482 if (!done)
2483 return 0;
2484
Simon Horman4a741432013-02-23 15:35:38 +09002485 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002486 "HTTP content check could not find a response body");
2487 return 1;
2488 }
2489
2490 /* Check that response body is not empty... */
2491 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002492 if (!done)
2493 return 0;
2494
Simon Horman4a741432013-02-23 15:35:38 +09002495 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002496 "HTTP content check found empty response body");
2497 return 1;
2498 }
2499
2500 /* Check the response content against the supplied string
2501 * or regex... */
2502 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2503 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2504 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002505 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002506
2507 /* if we don't match, we may need to wait more */
2508 if (!ret && !done)
2509 return 0;
2510
2511 if (ret) {
2512 /* content matched */
2513 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002514 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002515 "HTTP check matched unwanted content");
2516 else
Simon Horman4a741432013-02-23 15:35:38 +09002517 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002518 "HTTP content check matched");
2519 }
2520 else {
2521 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002522 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002523 "HTTP check did not match unwanted content");
2524 else
Simon Horman4a741432013-02-23 15:35:38 +09002525 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002526 "HTTP content check did not match");
2527 }
2528 break;
2529 }
2530 return 1;
2531}
2532
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002533/*
2534 * return the id of a step in a send/expect session
2535 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002536static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002537{
2538 struct tcpcheck_rule *cur = NULL, *next = NULL;
2539 int i = 0;
2540
Willy Tarreau213c6782014-10-02 14:51:02 +02002541 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002542 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002543 return 0;
2544
Simon Hormane16c1b32015-01-30 11:22:57 +09002545 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002546
2547 /* no step => first step */
2548 if (cur == NULL)
2549 return 1;
2550
2551 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002552 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002553 if (next->list.p == &cur->list)
2554 break;
2555 ++i;
2556 }
2557
2558 return i;
2559}
2560
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002561/*
2562 * return the latest known comment before (including) the given stepid
2563 * returns NULL if no comment found
2564 */
2565static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2566{
2567 struct tcpcheck_rule *cur = NULL;
2568 char *ret = NULL;
2569 int i = 0;
2570
2571 /* not even started anything yet, return latest comment found before any action */
2572 if (!check->current_step) {
2573 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2574 if (cur->action == TCPCHK_ACT_COMMENT)
2575 ret = cur->comment;
2576 else
2577 goto return_comment;
2578 }
2579 }
2580
2581 i = 1;
2582 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2583 if (cur->comment)
2584 ret = cur->comment;
2585
2586 if (i >= stepid)
2587 goto return_comment;
2588
2589 ++i;
2590 }
2591
2592 return_comment:
2593 return ret;
2594}
2595
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002596static void tcpcheck_main(struct connection *conn)
2597{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002598 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002599 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002600 int done = 0, ret = 0, step = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002601 struct check *check = conn->owner;
2602 struct server *s = check->server;
2603 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002604 struct list *head = check->tcpcheck_rules;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002605
Willy Tarreauef953952014-10-02 14:30:14 +02002606 /* here, we know that the check is complete or that it failed */
2607 if (check->result != CHK_RES_UNKNOWN)
2608 goto out_end_tcpcheck;
2609
2610 /* We have 4 possibilities here :
2611 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
2612 * connection attempt was made yet ;
2613 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2614 * does not exist (no rule), so a connection attempt was made
2615 * before coming here.
2616 * 3. we're coming back after having started with step 1, so we may
2617 * be waiting for a connection attempt to complete.
2618 * 4. the connection + handshake are complete
2619 *
2620 * #2 and #3 are quite similar, we want both the connection and the
2621 * handshake to complete before going any further. Thus we must always
2622 * wait for a connection to complete unless we're before and existing
2623 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002624 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002625
2626 /* find first rule and skip comments */
2627 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2628 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2629 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2630
Willy Tarreauef953952014-10-02 14:30:14 +02002631 if ((!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE)) &&
Willy Tarreau449f9522015-05-13 15:39:48 +02002632 (check->current_step || &next->list == head)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002633 /* we allow up to min(inter, timeout.connect) for a connection
2634 * to establish but only when timeout.check is set
2635 * as it may be to short for a full check otherwise
2636 */
2637 while (tick_is_expired(t->expire, now_ms)) {
2638 int t_con;
2639
2640 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2641 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2642
2643 if (s->proxy->timeout.check)
2644 t->expire = tick_first(t->expire, t_con);
2645 }
2646 return;
2647 }
2648
Willy Tarreauef953952014-10-02 14:30:14 +02002649 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002650 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002651 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002652 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002653 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002654
Willy Tarreau213c6782014-10-02 14:51:02 +02002655 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002656 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002657 check->last_started_step = NULL;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002658 check->bo->p = check->bo->data;
2659 check->bo->o = 0;
2660 check->bi->p = check->bi->data;
2661 check->bi->i = 0;
Willy Tarreau449f9522015-05-13 15:39:48 +02002662 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002663 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2664 if (s->proxy->timeout.check)
2665 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2666 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002667
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002668 /* It's only the rules which will enable send/recv */
2669 __conn_data_stop_both(conn);
2670
Willy Tarreauabca5b62013-12-06 14:19:25 +01002671 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002672 /* We have to try to flush the output buffer before reading, at
2673 * the end, or if we're about to send a string that does not fit
2674 * in the remaining space. That explains why we break out of the
2675 * loop after this control.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002676 */
2677 if (check->bo->o &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002678 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002679 check->current_step->action != TCPCHK_ACT_SEND ||
2680 check->current_step->string_len >= buffer_total_space(check->bo))) {
2681
Willy Tarreau1049b1f2014-02-02 01:51:17 +01002682 if (conn->xprt->snd_buf(conn, check->bo, 0) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002683 if (conn->flags & CO_FL_ERROR) {
2684 chk_report_conn_err(conn, errno, 0);
2685 __conn_data_stop_both(conn);
2686 goto out_end_tcpcheck;
2687 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002688 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002689 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002690 }
2691
Willy Tarreau263013d2015-05-13 11:59:14 +02002692 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002693 break;
2694
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002695 /* have 'next' point to the next rule or NULL if we're on the
2696 * last one, connect() needs this.
2697 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002698 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002699
2700 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002701 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002702 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002703
2704 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002705 if (&next->list == head)
2706 next = NULL;
2707
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002708 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2709 struct protocol *proto;
2710 struct xprt_ops *xprt;
2711
2712 /* mark the step as started */
2713 check->last_started_step = check->current_step;
2714 /* first, shut existing connection */
2715 conn_force_close(conn);
2716
2717 /* prepare new connection */
2718 /* initialization */
2719 conn_init(conn);
2720 conn_attach(conn, check, &check_conn_cb);
2721 conn->target = &s->obj_type;
2722
2723 /* no client address */
2724 clear_addr(&conn->addr.from);
2725
Simon Horman41f58762015-01-30 11:22:56 +09002726 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002727 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002728 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002729 }
2730 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002731 /* we'll connect to the addr on the server */
2732 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002733 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002734 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002735
2736 /* port */
2737 if (check->current_step->port)
2738 set_host_port(&conn->addr.to, check->current_step->port);
2739 else if (check->port)
2740 set_host_port(&conn->addr.to, check->port);
2741
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002742 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002743 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002744 }
2745 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002746 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002747 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002748 conn_prepare(conn, proto, xprt);
2749
Willy Tarreaue7dff022015-04-03 01:14:29 +02002750 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002751 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002752 ret = proto->connect(conn,
2753 1 /* I/O polling is always needed */,
2754 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002755 conn->flags |= CO_FL_WAKE_DATA;
2756 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2757 conn->send_proxy_ofs = 1;
2758 conn->flags |= CO_FL_SEND_PROXY;
2759 }
2760
2761 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002762 * - SF_ERR_NONE if everything's OK
2763 * - SF_ERR_SRVTO if there are no more servers
2764 * - SF_ERR_SRVCL if the connection was refused by the server
2765 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2766 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2767 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002768 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002769 * Note that we try to prevent the network stack from sending the ACK during the
2770 * connect() when a pure TCP check is used (without PROXY protocol).
2771 */
2772 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002773 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002774 /* we allow up to min(inter, timeout.connect) for a connection
2775 * to establish but only when timeout.check is set
2776 * as it may be to short for a full check otherwise
2777 */
2778 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2779
2780 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2781 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2782 t->expire = tick_first(t->expire, t_con);
2783 }
2784 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002785 case SF_ERR_SRVTO: /* ETIMEDOUT */
2786 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002787 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002788 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002789 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002790 comment = tcpcheck_get_step_comment(check, step);
2791 if (comment)
2792 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002793 set_server_check_status(check, HCHK_STATUS_L4CON, trash.str);
2794 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002795 case SF_ERR_PRXCOND:
2796 case SF_ERR_RESOURCE:
2797 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002798 step = tcpcheck_get_step_id(check);
2799 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002800 comment = tcpcheck_get_step_comment(check, step);
2801 if (comment)
2802 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002803 set_server_check_status(check, HCHK_STATUS_SOCKERR, trash.str);
2804 goto out_end_tcpcheck;
2805 }
2806
2807 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002808 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002809
2810 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002811 while (&check->current_step->list != head &&
2812 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002813 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002814
Willy Tarreauf2c87352015-05-13 12:08:21 +02002815 if (&check->current_step->list == head)
2816 break;
2817
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002818 /* don't do anything until the connection is established */
2819 if (!(conn->flags & CO_FL_CONNECTED)) {
2820 /* update expire time, should be done by process_chk */
2821 /* we allow up to min(inter, timeout.connect) for a connection
2822 * to establish but only when timeout.check is set
2823 * as it may be to short for a full check otherwise
2824 */
2825 while (tick_is_expired(t->expire, now_ms)) {
2826 int t_con;
2827
2828 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2829 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2830
2831 if (s->proxy->timeout.check)
2832 t->expire = tick_first(t->expire, t_con);
2833 }
2834 return;
2835 }
2836
2837 } /* end 'connect' */
2838 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2839 /* mark the step as started */
2840 check->last_started_step = check->current_step;
2841
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002842 /* reset the read buffer */
2843 if (*check->bi->data != '\0') {
2844 *check->bi->data = '\0';
2845 check->bi->i = 0;
2846 }
2847
2848 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
2849 conn->flags |= CO_FL_ERROR;
2850 chk_report_conn_err(conn, 0, 0);
2851 goto out_end_tcpcheck;
2852 }
2853
Willy Tarreauabca5b62013-12-06 14:19:25 +01002854 if (check->current_step->string_len >= check->bo->size) {
2855 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%d) at step %d",
2856 check->current_step->string_len, check->bo->size,
Simon Hormane16c1b32015-01-30 11:22:57 +09002857 tcpcheck_get_step_id(check));
Willy Tarreauabca5b62013-12-06 14:19:25 +01002858 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2859 goto out_end_tcpcheck;
2860 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002861
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002862 /* do not try to send if there is no space */
2863 if (check->current_step->string_len >= buffer_total_space(check->bo))
2864 continue;
2865
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002866 bo_putblk(check->bo, check->current_step->string, check->current_step->string_len);
2867 *check->bo->p = '\0'; /* to make gdb output easier to read */
2868
Willy Tarreauabca5b62013-12-06 14:19:25 +01002869 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002870 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002871
2872 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002873 while (&check->current_step->list != head &&
2874 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002875 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002876
2877 if (&check->current_step->list == head)
2878 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002879 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002880 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002881 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002882 goto out_end_tcpcheck;
2883
Willy Tarreau310987a2014-01-22 19:46:33 +01002884 if (conn->xprt->rcv_buf(conn, check->bi, check->bi->size) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002885 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
2886 done = 1;
2887 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
2888 /* Report network errors only if we got no other data. Otherwise
2889 * we'll let the upper layers decide whether the response is OK
2890 * or not. It is very common that an RST sent by the server is
2891 * reported as an error just after the last data chunk.
2892 */
2893 chk_report_conn_err(conn, errno, 0);
2894 goto out_end_tcpcheck;
2895 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002896 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002897 else
Willy Tarreau263013d2015-05-13 11:59:14 +02002898 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002899 }
2900
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002901 /* mark the step as started */
2902 check->last_started_step = check->current_step;
2903
2904
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002905 /* Intermediate or complete response received.
2906 * Terminate string in check->bi->data buffer.
2907 */
2908 if (check->bi->i < check->bi->size) {
2909 check->bi->data[check->bi->i] = '\0';
2910 }
2911 else {
2912 check->bi->data[check->bi->i - 1] = '\0';
2913 done = 1; /* buffer full, don't wait for more data */
2914 }
2915
2916 contentptr = check->bi->data;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002917
2918 /* Check that response body is not empty... */
Willy Tarreauec6b0122014-05-13 17:57:29 +02002919 if (!check->bi->i) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002920 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002921 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002922
2923 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002924 step = tcpcheck_get_step_id(check);
2925 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002926 comment = tcpcheck_get_step_comment(check, step);
2927 if (comment)
2928 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002929 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2930
2931 goto out_end_tcpcheck;
2932 }
2933
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002934 if (!done && (check->current_step->string != NULL) && (check->bi->i < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002935 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002936
Willy Tarreaua970c282013-12-06 12:47:19 +01002937 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002938 if (check->current_step->string != NULL)
2939 ret = my_memmem(contentptr, check->bi->i, check->current_step->string, check->current_step->string_len) != NULL;
2940 else if (check->current_step->expect_regex != NULL)
2941 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002942
2943 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002944 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002945
2946 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002947 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002948 if (ret) {
2949 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002950 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002951 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002952 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002953 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002954 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002955 }
2956 else {
2957 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002958 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002959 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002960 comment = tcpcheck_get_step_comment(check, step);
2961 if (comment)
2962 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002963 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2964 goto out_end_tcpcheck;
2965 }
2966 /* matched and was supposed to => OK, next step */
2967 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002968 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002969 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002970
2971 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002972 while (&check->current_step->list != head &&
2973 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002974 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002975
Willy Tarreauf2c87352015-05-13 12:08:21 +02002976 if (&check->current_step->list == head)
2977 break;
2978
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002979 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002980 goto tcpcheck_expect;
2981 __conn_data_stop_recv(conn);
2982 }
2983 }
2984 else {
2985 /* not matched */
2986 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002987 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002988 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002989 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002990
2991 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002992 while (&check->current_step->list != head &&
2993 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002994 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002995
Willy Tarreauf2c87352015-05-13 12:08:21 +02002996 if (&check->current_step->list == head)
2997 break;
2998
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002999 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003000 goto tcpcheck_expect;
3001 __conn_data_stop_recv(conn);
3002 }
3003 /* not matched but was supposed to => ERROR */
3004 else {
3005 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003006 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003007 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003008 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003009 }
3010 else {
3011 /* we were looking for a regex */
3012 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003013 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003014 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003015 comment = tcpcheck_get_step_comment(check, step);
3016 if (comment)
3017 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003018 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
3019 goto out_end_tcpcheck;
3020 }
3021 }
3022 } /* end expect */
3023 } /* end loop over double chained step list */
3024
Willy Tarreau263013d2015-05-13 11:59:14 +02003025 /* We're waiting for some I/O to complete, we've reached the end of the
3026 * rules, or both. Do what we have to do, otherwise we're done.
3027 */
3028 if (&check->current_step->list == head && !check->bo->o) {
3029 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3030 goto out_end_tcpcheck;
3031 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003032
Willy Tarreau53c5a042015-05-13 11:38:17 +02003033 /* warning, current_step may now point to the head */
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003034 if (check->bo->o)
3035 __conn_data_want_send(conn);
3036
Willy Tarreau53c5a042015-05-13 11:38:17 +02003037 if (&check->current_step->list != head &&
3038 check->current_step->action == TCPCHK_ACT_EXPECT)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003039 __conn_data_want_recv(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003040 return;
3041
3042 out_end_tcpcheck:
3043 /* collect possible new errors */
3044 if (conn->flags & CO_FL_ERROR)
3045 chk_report_conn_err(conn, 0, 0);
3046
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003047 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003048 check->current_step = NULL;
3049
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003050 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003051 conn->flags |= CO_FL_ERROR;
3052
3053 __conn_data_stop_both(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003054 return;
3055}
3056
Simon Hormanb1900d52015-01-30 11:22:54 +09003057const char *init_check(struct check *check, int type)
3058{
3059 check->type = type;
3060
3061 /* Allocate buffer for requests... */
3062 if ((check->bi = calloc(sizeof(struct buffer) + global.tune.chksize, sizeof(char))) == NULL) {
3063 return "out of memory while allocating check buffer";
3064 }
3065 check->bi->size = global.tune.chksize;
3066
3067 /* Allocate buffer for responses... */
3068 if ((check->bo = calloc(sizeof(struct buffer) + global.tune.chksize, sizeof(char))) == NULL) {
3069 return "out of memory while allocating check buffer";
3070 }
3071 check->bo->size = global.tune.chksize;
3072
3073 /* Allocate buffer for partial results... */
3074 if ((check->conn = calloc(1, sizeof(struct connection))) == NULL) {
3075 return "out of memory while allocating check connection";
3076 }
3077
3078 check->conn->t.sock.fd = -1; /* no agent in progress yet */
3079
3080 return NULL;
3081}
3082
Simon Hormanbfb5d332015-01-30 11:22:55 +09003083void free_check(struct check *check)
3084{
3085 free(check->bi);
3086 free(check->bo);
3087 free(check->conn);
3088}
3089
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003090void email_alert_free(struct email_alert *alert)
3091{
3092 struct tcpcheck_rule *rule, *back;
3093
3094 if (!alert)
3095 return;
3096
3097 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list)
3098 free(rule);
3099 free(alert);
3100}
3101
3102static struct task *process_email_alert(struct task *t)
3103{
3104 struct check *check = t->context;
3105 struct email_alertq *q;
3106
3107 q = container_of(check, typeof(*q), check);
3108
3109 if (!(check->state & CHK_ST_ENABLED)) {
3110 if (LIST_ISEMPTY(&q->email_alerts)) {
3111 /* All alerts processed, delete check */
3112 task_delete(t);
3113 task_free(t);
3114 check->task = NULL;
3115 return NULL;
3116 } else {
3117 struct email_alert *alert;
3118
3119 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
3120 check->tcpcheck_rules = &alert->tcpcheck_rules;
3121 LIST_DEL(&alert->list);
3122
3123 check->state |= CHK_ST_ENABLED;
3124 }
3125
3126 }
3127
3128 process_chk(t);
3129
3130 if (!(check->state & CHK_ST_INPROGRESS) && check->tcpcheck_rules) {
3131 struct email_alert *alert;
3132
3133 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3134 email_alert_free(alert);
3135
3136 check->tcpcheck_rules = NULL;
3137 check->state &= ~CHK_ST_ENABLED;
3138 }
3139 return t;
3140}
3141
3142static int init_email_alert_checks(struct server *s)
3143{
3144 int i;
3145 struct mailer *mailer;
3146 const char *err_str;
3147 struct proxy *p = s->proxy;
3148
3149 if (p->email_alert.queues)
3150 /* Already initialised, nothing to do */
3151 return 1;
3152
3153 p->email_alert.queues = calloc(p->email_alert.mailers.m->count, sizeof *p->email_alert.queues);
3154 if (!p->email_alert.queues) {
3155 err_str = "out of memory while allocating checks array";
3156 goto error_alert;
3157 }
3158
3159 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3160 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
3161 struct email_alertq *q = &p->email_alert.queues[i];
3162 struct check *check = &q->check;
3163
3164
3165 LIST_INIT(&q->email_alerts);
3166
Pieter Baauw235fcfc2016-02-13 15:33:40 +01003167 check->inter = p->email_alert.mailers.m->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003168 check->rise = DEF_AGENT_RISETIME;
3169 check->fall = DEF_AGENT_FALLTIME;
3170 err_str = init_check(check, PR_O2_TCPCHK_CHK);
3171 if (err_str) {
3172 goto error_free;
3173 }
3174
3175 check->xprt = mailer->xprt;
3176 if (!get_host_port(&mailer->addr))
3177 /* Default to submission port */
3178 check->port = 587;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003179 check->addr = mailer->addr;
3180 check->server = s;
3181 }
3182
3183 return 1;
3184
3185error_free:
3186 while (i-- > 1)
3187 task_free(p->email_alert.queues[i].check.task);
3188 free(p->email_alert.queues);
3189 p->email_alert.queues = NULL;
3190error_alert:
3191 Alert("Email alert [%s] could not be initialised: %s\n", p->id, err_str);
3192 return 0;
3193}
3194
3195
3196static int add_tcpcheck_expect_str(struct list *list, const char *str)
3197{
3198 struct tcpcheck_rule *tcpcheck;
3199
3200 tcpcheck = calloc(1, sizeof *tcpcheck);
3201 if (!tcpcheck)
3202 return 0;
3203
3204 tcpcheck->action = TCPCHK_ACT_EXPECT;
3205 tcpcheck->string = strdup(str);
3206 if (!tcpcheck->string) {
3207 free(tcpcheck);
3208 return 0;
3209 }
3210
3211 LIST_ADDQ(list, &tcpcheck->list);
3212 return 1;
3213}
3214
3215static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3216{
3217 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003218 const char *in;
3219 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003220 int i;
3221
3222 tcpcheck = calloc(1, sizeof *tcpcheck);
3223 if (!tcpcheck)
3224 return 0;
3225
3226 tcpcheck->action = TCPCHK_ACT_SEND;
3227
3228 tcpcheck->string_len = 0;
3229 for (i = 0; strs[i]; i++)
3230 tcpcheck->string_len += strlen(strs[i]);
3231
3232 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3233 if (!tcpcheck->string) {
3234 free(tcpcheck);
3235 return 0;
3236 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003237
Willy Tarreau64345aa2016-08-10 19:29:09 +02003238 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003239 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003240 for (in = strs[i]; (*dst = *in++); dst++);
3241 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003242
3243 LIST_ADDQ(list, &tcpcheck->list);
3244 return 1;
3245}
3246
3247static int enqueue_one_email_alert(struct email_alertq *q, const char *msg)
3248{
3249 struct email_alert *alert = NULL;
3250 struct tcpcheck_rule *tcpcheck;
3251 struct check *check = &q->check;
3252 struct proxy *p = check->server->proxy;
3253
3254 alert = calloc(1, sizeof *alert);
3255 if (!alert) {
3256 goto error;
3257 }
3258 LIST_INIT(&alert->tcpcheck_rules);
3259
3260 tcpcheck = calloc(1, sizeof *tcpcheck);
3261 if (!tcpcheck)
3262 goto error;
3263 tcpcheck->action = TCPCHK_ACT_CONNECT;
3264 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3265
3266 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3267 goto error;
3268
3269 {
3270 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3271 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3272 goto error;
3273 }
3274
3275 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3276 goto error;
3277
3278 {
3279 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3280 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3281 goto error;
3282 }
3283
3284 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3285 goto error;
3286
3287 {
3288 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3289 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3290 goto error;
3291 }
3292
3293 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3294 goto error;
3295
3296 {
3297 const char * const strs[2] = { "DATA\r\n" };
3298 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3299 goto error;
3300 }
3301
3302 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3303 goto error;
3304
3305 {
3306 struct tm tm;
3307 char datestr[48];
3308 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003309 "From: ", p->email_alert.from, "\r\n",
3310 "To: ", p->email_alert.to, "\r\n",
3311 "Date: ", datestr, "\r\n",
3312 "Subject: [HAproxy Alert] ", msg, "\r\n",
3313 "\r\n",
3314 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003315 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003316 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003317 NULL
3318 };
3319
3320 get_localtime(date.tv_sec, &tm);
3321
3322 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3323 goto error;
3324 }
3325
3326 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3327 goto error;
3328 }
3329
3330 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3331 goto error;
3332
3333 {
3334 const char * const strs[2] = { "QUIT\r\n" };
3335 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3336 goto error;
3337 }
3338
3339 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3340 goto error;
3341
3342 if (!check->task) {
3343 struct task *t;
3344
3345 if ((t = task_new()) == NULL)
3346 goto error;
3347
3348 check->task = t;
3349 t->process = process_email_alert;
3350 t->context = check;
3351
3352 /* check this in one ms */
3353 t->expire = tick_add(now_ms, MS_TO_TICKS(1));
3354 check->start = now;
3355 task_queue(t);
3356 }
3357
3358 LIST_ADDQ(&q->email_alerts, &alert->list);
3359
3360 return 1;
3361
3362error:
3363 email_alert_free(alert);
3364 return 0;
3365}
3366
3367static void enqueue_email_alert(struct proxy *p, const char *msg)
3368{
3369 int i;
3370 struct mailer *mailer;
3371
3372 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3373 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
3374 if (!enqueue_one_email_alert(&p->email_alert.queues[i], msg)) {
3375 Alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
3376 return;
3377 }
3378 }
3379
3380 return;
3381}
3382
3383/*
3384 * Send email alert if configured.
3385 */
Simon Horman64e34162015-02-06 11:11:57 +09003386void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003387{
3388 va_list argp;
3389 char buf[1024];
3390 int len;
3391 struct proxy *p = s->proxy;
3392
Simon Horman64e34162015-02-06 11:11:57 +09003393 if (!p->email_alert.mailers.m || level > p->email_alert.level ||
3394 format == NULL || !init_email_alert_checks(s))
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003395 return;
3396
3397 va_start(argp, format);
3398 len = vsnprintf(buf, sizeof(buf), format, argp);
3399 va_end(argp);
3400
3401 if (len < 0) {
Cyril Bontéb65e0332015-12-04 03:07:08 +01003402 Alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003403 return;
3404 }
3405
3406 enqueue_email_alert(p, buf);
3407}
3408
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003409/*
3410 * Return value:
3411 * the port to be used for the health check
3412 * 0 in case no port could be found for the check
3413 */
3414int srv_check_healthcheck_port(struct check *chk)
3415{
3416 int i = 0;
3417 struct server *srv = NULL;
3418
3419 srv = chk->server;
3420
3421 /* If neither a port nor an addr was specified and no check transport
3422 * layer is forced, then the transport layer used by the checks is the
3423 * same as for the production traffic. Otherwise we use raw_sock by
3424 * default, unless one is specified.
3425 */
3426 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003427 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003428 chk->send_proxy |= (srv->pp_opts);
3429 }
3430
3431 /* by default, we use the health check port ocnfigured */
3432 if (chk->port > 0)
3433 return chk->port;
3434
3435 /* try to get the port from check_core.addr if check.port not set */
3436 i = get_host_port(&chk->addr);
3437 if (i > 0)
3438 return i;
3439
3440 /* try to get the port from server address */
3441 /* prevent MAPPORTS from working at this point, since checks could
3442 * not be performed in such case (MAPPORTS impose a relative ports
3443 * based on live traffic)
3444 */
3445 if (srv->flags & SRV_F_MAPPORTS)
3446 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003447
3448 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003449 if (i > 0)
3450 return i;
3451
3452 return 0;
3453}
3454
Willy Tarreau865c5142016-12-21 20:04:48 +01003455__attribute__((constructor))
3456static void __check_init(void)
3457{
3458 hap_register_post_check(start_checks);
3459}
3460
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003461
Willy Tarreaubd741542010-03-16 18:46:54 +01003462/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003463 * Local variables:
3464 * c-indent-level: 8
3465 * c-basic-offset: 8
3466 * End:
3467 */